WEBVTT
00:00:02.479 --> 00:00:20.879
As Agentic AI became more powerful and token hungry in the first months of 2026, the AI decision makers a month or so ago started to collectively tighten their token maxing belts to prepare for a more expensive AI future.
00:00:21.039 --> 00:00:21.359
Right?
00:00:21.519 --> 00:00:26.559
It's the shift from token maxing to token efficiency we've been talking about here for months.
00:00:26.800 --> 00:00:33.520
But then something drastic and kind of unexpected happened that I don't think enough people are talking about.
00:00:33.759 --> 00:00:50.159
OpenAI on Thursday revealed that its GPT-5.6 soul model improved itself and actually improved itself so much after its initial release that they were cutting prices on some of the variants of the model by 80%.
00:00:51.039 --> 00:00:53.200
So we've reached kind of a new juncture.
00:00:53.359 --> 00:01:02.079
Today's Frontier models are not just helping to create or distill their smaller variants, but they're actually improving themselves after release.
00:01:02.320 --> 00:01:09.519
Hence the next AI buzzword you'll be hearing a lot over the next year: recursive self-improvement or RSI.
00:01:09.840 --> 00:01:20.159
So, in short, recursive self-improvement is a process where an artificial intelligence system uses its own capabilities to design, code, and build a more advanced version or a better version of itself.
00:01:20.319 --> 00:01:33.439
And you'll be hearing this a lot more recently because RSI has become just the dominant tech headline over the past few weeks with a handful of events that we're gonna break down on today's show.
00:01:33.599 --> 00:01:48.959
But unlike so many buzzwords that we've covered here on everyday AI over the years, RSI is one that actually today's AI implementations are gonna be impacted by, and so too will tomorrow's AI security.
00:01:49.200 --> 00:02:02.400
In practical sense, like in OpenAI's case, RSI can change what AI capabilities companies everywhere can actually afford, but it also sheds light on potential downsides of self-improving models, doubling down the wrong path.
00:02:02.640 --> 00:02:08.719
So, what the heck is RSI and why is every recent AI story touching on it?
00:02:08.800 --> 00:02:10.319
And what do we all need to know?
00:02:10.479 --> 00:02:13.039
Well, let's jump straight into the big picture.
00:02:13.280 --> 00:02:18.400
The big picture here is AI has started building better and more efficient AI.
00:02:18.560 --> 00:02:24.800
So Frontier AI models now improve themselves, they're rewriting their own code and training their smaller siblings.
00:02:24.960 --> 00:02:34.159
So this is different than when you know Anthropic and OpenAI are talking about using their own models to build products like Claude Code or like Codex.
00:02:34.479 --> 00:02:46.159
This is when the models are actually making themselves better, and that process is called RSI, and it's no longer some future science fiction, it is a near-term reality.
00:02:46.319 --> 00:02:56.719
So OpenAI, as an example, says that its GPT-5.6 soul model made its Luna model, the smaller variant, just cheaper and more efficient.
00:02:56.879 --> 00:02:58.719
And then they cut prices by 80%.
00:02:59.520 --> 00:03:05.039
But the full human-free recursive self-improvement hasn't yet arrived.
00:03:05.280 --> 00:03:19.680
But over the past two months, there's a lot of AI news stories that we're gonna be connecting the dots on uh the dots on today that show that, well, this probably isn't like a 2029-2028 thing that a lot of people have been saying.
00:03:19.840 --> 00:03:26.159
It's probably a 2027 thing, and that changes things maybe in both a good and a scary way.
00:03:26.560 --> 00:03:28.639
So, on today's show, here's where you're gonna learn.
00:03:28.800 --> 00:03:37.120
You're gonna learn why OpenAI was able to cut prices up to 80% when everyone predicted that AI costs would continue to climb in 2026.
00:03:37.199 --> 00:03:44.479
You're gonna know more about this eight eight-week timeline that turned RSI from research jargon into mainstream headlines.
00:03:44.639 --> 00:03:55.520
You're gonna understand by the end of today's show what RSI actually means in plain English and what Sam Altman, the CEO of OpenAI's singularity claim that he recently made actually means.
00:03:55.680 --> 00:04:06.639
And you're gonna know why the people building RSI are actually asking for maybe the industry to pause a little bit and pace itself and how your business should respond to all of this.
00:04:06.879 --> 00:04:07.919
All right, let's get into it.
00:04:08.000 --> 00:04:09.199
Welcome to Everyday AI.
00:04:09.360 --> 00:04:11.759
My name is Jordan Malton, and well, this thing's for you.
00:04:11.919 --> 00:04:18.319
This is your daily unedited, unscripted live stream podcast, and free daily newsletter helping business leaders like you and me keep up.
00:04:18.399 --> 00:04:19.839
I do all the hard research.
00:04:20.000 --> 00:04:25.600
So you sit back, listen, read our newsletter, enjoy the benefits, and you grow your company and career.
00:04:25.759 --> 00:04:31.279
So it starts here, but make sure to subscribe to our free daily newsletter at your everydayai.com.
00:04:31.360 --> 00:04:36.560
We're going to be recapping all the highlights from today's show as well as all of the other AI news that you need to know.
00:04:36.879 --> 00:04:46.160
So let's get into the latest uh three to four letter acronym in our ever-evolving bowl of alphabet AI soup.
00:04:46.240 --> 00:04:47.759
We are talking RSI.
00:04:48.720 --> 00:04:53.199
So let's talk about the last uh two months.
00:04:53.519 --> 00:05:03.439
Because I mean, yes, I've been talking about uh recursive self-improvement on this show probably for the last two or so years, since I've been doing this for the last three and a half years.
00:05:03.600 --> 00:05:09.600
Um, as it seemed like recursive self-improvement was actually going to be something that happened this decade.
00:05:09.680 --> 00:05:10.319
Um, right.
00:05:10.399 --> 00:05:18.399
Because if you when I started this thing in 2023, most people thought that recursive self-improvement was at best a 2030 thing.
00:05:18.560 --> 00:05:32.000
Yet here we are, where we're already getting glimpses and sniffs of real RSI, and probably we will get full RSI by next year, which is actually crazy to think about.
00:05:32.160 --> 00:05:38.639
So, but let's kind of uh look at the last two months and some of the different uh dots that we're gonna connect here.
00:05:38.720 --> 00:05:44.959
So in June, Anthropic published a post on its website called When AI Builds Itself.
00:05:45.120 --> 00:05:55.120
Uh, also in June, Google researchers mapped four different routes from AGI to superintelligence, and one of those routes well relied heavily on recursive self-improvement.
00:05:55.279 --> 00:06:04.800
Uh, then in July, uh OpenAI launched an internal RSI benchmark, uh, measuring different models' ability to well improve themselves.
00:06:04.959 --> 00:06:11.759
Then, like I said, its soul model post-trained its smaller sibling Luna at the end of July.
00:06:11.920 --> 00:06:18.639
And then also in a podcast interview, uh, OpenAI CEO Sam Altman declared that we are in the singularity.
00:06:18.720 --> 00:06:34.639
Uh yeah, and the news cycle was just warming up because uh at the end of July, so last week, startup recursive superintelligence signed a$400 million AWS compute deal to automate AI research.
00:06:34.720 --> 00:06:38.879
So not only is it the labs that all of us know, right?
00:06:38.959 --> 00:06:50.879
Kind of like the big four plus meta and acts, all working toward, you know, some version of recursive self-improvement, but it's also, you know, new startups that are coming in heavily funded just to do that and nothing else.
00:06:51.040 --> 00:07:26.399
And on that same day, so on July 28th, over a thousand, and I think that number is like 1300 now, Frontier Lab employees published a paper called Pacing the Frontier, which essentially is kind of asking the governments, uh, the US governments to maybe slow down or at least prepare for the possibility that there needs to be some international um kind of agreement when it comes to the pace of AI, because a lot of people signing that letter in the comments said that, well, AI is developing maybe too quickly and that society can't keep up.
00:07:26.639 --> 00:07:33.519
And let me just say this right now, um, and I need to hit a pause because I think that, you know, I am not in the Silicon Valley bubble, right?
00:07:33.600 --> 00:07:38.000
I don't work at a big tech company, um, but I talk about AI every single day.
00:07:38.079 --> 00:07:43.120
And I do talk to those people building uh AI very recently uh very frequently.
00:07:43.360 --> 00:07:48.079
And if you're listening to the show, you're probably uh, you know, more like me, right?
00:07:48.240 --> 00:07:53.600
I know we have a lot of people um at those big tech companies, but um I'll say this.
00:07:54.560 --> 00:08:02.959
The rest of the world right now, when they're viewing artificial intelligence, right, most people are saying, oh yeah, I use Copilot to make my emails better.
00:08:03.120 --> 00:08:08.720
Or they're saying, you know, oh, I use, you know, uh Chat GPT now instead of Google, uh, right.
00:08:08.879 --> 00:08:14.720
Or they're saying, oh, I use, you know, Google Gemini to, you know, make, you know, cool graphics, right?
00:08:15.120 --> 00:08:16.000
That's it.
00:08:16.240 --> 00:08:20.240
Um, so I would say that, you know, aside from our audience, right?
00:08:20.399 --> 00:08:24.000
I want you to think of yourself right now as you're kind of in that AI bubble, right?
00:08:24.079 --> 00:08:28.399
Especially if you're a frequent listener or if you've been listening for a long time.
00:08:28.560 --> 00:08:32.000
You know, we're kind of in that bubble, even though we're not in Silicon Valley, right?
00:08:32.080 --> 00:08:37.840
We understand how uh capable these models are, but the rest of the world really has no clue, right?
00:08:38.000 --> 00:08:46.720
All they're seeing is they're seeing, you know, oh, you know, all these AI models, you know, I'm just going in here, I'm finding the one I want, and you know, maybe my work's a little bit easier.
00:08:46.879 --> 00:08:53.279
Or, you know, I think that there's a lot of people now looking at AI, and there is this uh this recent anti-AI swing, right?
00:08:53.360 --> 00:09:00.080
When it comes to data centers and, you know, the college graduates that were booing, you know, commencement speakers who are talking about AI.
00:09:00.159 --> 00:09:06.159
So there's a lot of just bad information uh out there when it comes to AI that has created this anti-AI backlash.
00:09:06.320 --> 00:09:16.960
So I think, if nothing else, there is such a huge disconnect between the general uh, you know, non everyday AI listener type, right?
00:09:17.120 --> 00:09:23.600
Um, or you know, think of your in your organization, you know, there's probably a team of AI champions and then there's everyone else, right?
00:09:23.759 --> 00:09:34.799
So for I would say 90% of the US working population, uh, they really don't have any clue what today's AI systems are capable of, right?
00:09:34.960 --> 00:09:49.919
If you sat all those people down and showed them uh what something like, you know, GPT-5-6 soul or something like you know, Fable uh five and you know, in Codecs and Claude Code, if you show them what it could do, they would not believe you.
00:09:50.080 --> 00:09:51.759
They would say, oh, this is from the future, right?
00:09:51.840 --> 00:10:02.639
And I think that's why we got this letter from the Frontier Labs called Pacing the Frontier, uh, which we'll talk about here in a couple of minutes, that are saying, like, hey, we might need to pace uh the rate of AI acceleration.
00:10:02.799 --> 00:10:18.480
And then days later, uh, which is kind of a sounds like a small uh side plot, but it's not a formal OpenAI employee uh who became a thinking uh Thinking Labs co-founder rejoined OpenAI to specifically um work on RSI.
00:10:18.559 --> 00:10:22.399
And then Google also, right, all these things happening at once.
00:10:22.480 --> 00:10:25.679
Uh, you know, this was uh the end of July as well.
00:10:25.919 --> 00:10:30.559
Uh a Google exec uh called CapEx an RSI bet.
00:10:30.799 --> 00:10:31.840
So what does that mean, right?
00:10:31.919 --> 00:10:36.399
Uh collectively, the industry is spending hundreds of billions of dollars on data centers.
00:10:36.559 --> 00:10:41.360
And essentially a Google exec said, well, this is because of recursive self-improvement, right?
00:10:41.519 --> 00:10:44.159
You need all this compute to be able to do this, right?
00:10:44.240 --> 00:10:44.960
That's the difference.
00:10:45.120 --> 00:10:56.399
You know, when you have a team of researchers, uh, maybe, you know, three to four years ago, where this kind of the AI researcher wasn't yet automated or on its way to being automated, right?
00:10:56.559 --> 00:10:58.799
You just had to use a lot more people.
00:10:59.120 --> 00:10:59.919
Um, right.
00:11:00.080 --> 00:11:05.440
Yes, you were still able to, you know, the concept of uh recursive self-improvement is not new.
00:11:05.519 --> 00:11:14.639
It's been around since, you know, I would say like 2016, technically, with Alpha Go, um, you know, and also some of the earlier GPT models, uh, right.
00:11:14.720 --> 00:11:20.799
But that was much more kind of human hand holding, uh, you know, AI models to improve, right?
00:11:20.960 --> 00:11:26.080
Where now it's more of uh human oversight and the models are kind of doing a lot of the work on themselves.
00:11:26.159 --> 00:11:29.840
So, like I said, it's not full human-free, uh, but we are at the point now.
00:11:30.080 --> 00:11:44.080
But, you know, a high-ranking Google exec essentially just said last week that yeah, all this money that's being spent on these data centers, this this hundreds of billions of dollars, that's so that is for recursive self-improvement.
00:11:44.159 --> 00:11:56.559
So these models are just gonna be able to build not only the next version of themselves, but once that next version has been released, like OpenAI just did, well, it's gonna improve itself to hopefully make it cheaper, faster, better, right?
00:11:56.720 --> 00:12:05.519
But then also, I think, right, I've always been saying I'm a huge believer in this concept that eventually there's gonna be thousands of small models, uh, right.
00:12:05.679 --> 00:12:07.039
I think that that's gonna be the truth.
00:12:07.120 --> 00:12:12.320
There's gonna be thousands of models that help with marketing, there's gonna be thousands of models that help with uh medicine.
00:12:12.399 --> 00:12:14.960
There's right, there's gonna be thousands, but small ones.
00:12:15.039 --> 00:12:25.759
And these this is where I think the future is heading is recursive self-improvement, where uh these model makers are gonna be able to spin out, you know, probably the ones first that make the most sense to make money, right?
00:12:25.840 --> 00:12:30.000
And then from after that, I'm guessing we'll we're gonna equally see or hopefully see.
00:12:30.080 --> 00:12:37.840
It's my hope that we'll uh you know, equally see uh, you know, all of these small models that come from RSI that are gonna be able to do good in the world, right?
00:12:37.919 --> 00:12:44.000
And we're already seeing that uh with things like medicine and biology and you know, right now, uh defensive cybersecurity.
00:12:44.240 --> 00:12:51.840
So now let's talk about that Thursday shocker that kind of changed, I think, the near-term AI strategy for a lot of business owners out there.
00:12:52.000 --> 00:13:03.039
Uh so for months, right, we've been hearing this, this, this concept um that these powerful AI agents, well, they were too token hungry, you know.
00:13:03.200 --> 00:13:13.039
So I think in late 2025, kind of with the the advent of of Claude Code, Claude Cowork in early 2026, uh, and then codex in February, right?
00:13:13.120 --> 00:13:19.120
It kind of got to this point where everyone was spending as many tokens as possible because companies everywhere were like, oh my gosh, right?
00:13:19.200 --> 00:13:26.159
Those on the bleeding edge, those AI champions were like, my gosh, these systems can do absolutely anything, which was great, right?
00:13:26.240 --> 00:13:28.559
But the revenue didn't always directly follow.
00:13:28.720 --> 00:13:41.440
And so what that meant is while you know some companies were waiting to see the uh the fruits of their token maxing labor, they had to cut back, right, and focus a little bit more on token efficiency or as some people say, value maxing.
00:13:41.600 --> 00:13:47.759
So we kind of thought that at this point, well, okay, we're just gonna probably start spending less, right?
00:13:48.000 --> 00:13:59.759
But open AI came and threw that out the window because open AI said that Seoul, their GPD-5.6 soul, adapted an existing post-training setup for the smaller Luna model.
00:13:59.919 --> 00:14:04.000
So essentially, right, and they actually shared this, and we shared this in our newsletter as well.
00:14:04.159 --> 00:14:08.639
Uh, they kind of shared and they did this, I believe, all in codecs.
00:14:08.799 --> 00:14:14.559
Um, so it's it's it's you know, wasn't some far off, you know, science fiction.
00:14:14.639 --> 00:14:28.799
It's some uh uh according to someone that tweeted about this, um, it is some open AI researchers who used codecs and they used the GBT56 soul model to improve the smaller versions.
00:14:28.960 --> 00:14:31.440
Uh, so the medium size is GBT5.
00:14:35.759 --> 00:14:41.200
And then GBD56 Luna, which they were able to cut the prices on that by 80%, right?
00:14:41.440 --> 00:14:45.200
And the fact that that happened, and I still don't think enough people are talking about it, right?
00:14:45.360 --> 00:14:53.200
Because if you compare as an example, um, and I did talk about this on the show uh yesterday, maybe, or maybe it was Friday, um, right?
00:14:53.360 --> 00:15:06.879
But now you have GBD56 Luna, which according to artificial uh analysis is roughly about the same as uh Claude Sonnet 5, but it is 25x cheaper per task completion, right?
00:15:07.120 --> 00:15:15.440
So now it's almost like, wait, you know, there is this narrative in quarter two that we were gonna have to start using less AI.
00:15:15.600 --> 00:15:25.840
And now all of a sudden, because we are getting this hint of recursive self-improvement from open AI first, now we have to start rethinking that strategy immediately.
00:15:26.399 --> 00:15:28.240
So let's define it a little bit.
00:15:28.320 --> 00:15:28.960
Uh, all right.
00:15:29.120 --> 00:15:36.159
So uh RSI is essentially when an AI helps improve another AI, it could be a from a different company, right?
00:15:36.240 --> 00:15:40.080
It could be its own um smaller version, like we saw with the G V D5.
00:15:41.279 --> 00:15:43.600
Um uh working on G V D5.
00:15:44.720 --> 00:15:50.960
And we've seen from other companies talking about kind of their big model helping train some of their smaller models.
00:15:51.039 --> 00:15:54.320
Uh, and we've seen also this is similar to distillation, right?
00:15:54.480 --> 00:15:56.799
So um intercompany distillation.
00:15:56.960 --> 00:16:08.559
Um, you know, generally when they're talking about that, yes, there's humans involved, but I would say in 2026, uh, I I think the labs will start talking about this more as it becomes more commonplace.
00:16:08.720 --> 00:16:17.440
Um, but you know, now it's with these smaller models, it is very commonplace for companies to distill the smaller versions from the big versions.
00:16:17.600 --> 00:16:23.039
And I think that's also why we're sometimes seeing um the versions come out at a different time, right?
00:16:23.200 --> 00:16:26.559
So, as an example, Fable 5 came first, right?
00:16:26.720 --> 00:16:33.840
And then we or you know, Fable and Mythos 5 came first, and then we got you know Sonnet 5, and then we got Opus five.
00:16:33.919 --> 00:16:43.840
So uh, you know, Anthropic has kind of hinted in the past that a lot of their code, up to you know, 90%, I think according to some estimates, you know, 90% of their code is actually written by their own models.
00:16:44.399 --> 00:16:54.720
So uh, but the Google exec though, um, I think this is important, um, compared it to history, talking about steam engines were used to build the next steam engine.
00:16:54.879 --> 00:17:01.840
And I think to understand why that matters, uh, we have to take a look at the um all the other definitions, right?
00:17:01.919 --> 00:17:06.480
Um, and and how recursive self-improvement actually fits on the current timeline.
00:17:06.640 --> 00:17:22.240
Because for the most part, uh it it, at least if you've been following uh and keeping track at home, the um kind of the the path uh to um you know where AI is eventually going to go, whether you're excited about that or uh not.
00:17:22.480 --> 00:17:26.720
And you know, I constantly find myself going on on each side of that equation, right?
00:17:26.799 --> 00:17:35.599
But you think of, well, we're in the artificial intelligence um era, and it seems like most companies are focused on super intelligence.
00:17:35.839 --> 00:17:40.799
Okay, so superintelligence is when, well, all AI is smarter than all humans.
00:17:40.960 --> 00:17:43.839
Um, but AGI is kind of that stepping point.
00:17:44.000 --> 00:17:53.519
So uh, you know, if you had to say three things, it's AI to AGI, which is artificial general intelligence, um, and then you have superintelligence.
00:17:53.680 --> 00:18:03.680
And that's I think superintelligence, if you uh if superintelligence goes wrong, that's where you start getting into the Terminator and Skynet uh type comparisons, right?
00:18:03.759 --> 00:18:10.720
When superintelligence goes right, that's when, well, in theory, we could cure most diseases, right?
00:18:10.960 --> 00:18:14.559
Um, but there's those kind of steps belong uh along the way.
00:18:14.720 --> 00:18:26.240
Um, and I think that uh according to um a lot of people's timelines, um actually RSI is kind of what pushes from AGI to ASI.
00:18:26.480 --> 00:18:30.000
Uh and I know we're throwing a lot of um acronyms out there.
00:18:30.079 --> 00:18:34.799
So I'm in the firm belief that we've already hit artificial general intelligence, right?
00:18:34.960 --> 00:18:38.160
I know it's kind of a touchy topic, it's it's moving goalposts.
00:18:38.319 --> 00:18:41.440
Uh, luckily, you know, I've been saying this since uh last year.
00:18:41.519 --> 00:18:44.799
Luckily, uh NVIDIA CEO Jensen Wong uh said something similar.
00:18:44.880 --> 00:18:49.920
So I don't feel as crazy when you know super smart people start saying, yeah, we're probably past AGI, right?
00:18:50.000 --> 00:18:59.119
So artificial general intelligence is well when most AI systems can produce economically viable work at the same rate or higher than most humans.
00:18:59.200 --> 00:18:59.759
Uh, right.
00:18:59.839 --> 00:19:16.079
And I think as we've seen uh agentic AI be able to use computers, being able to browse the internet, being able to see and hear and listen and do all these things technically, you know, through a computer that an average human sitting in front of a computer could do, you know, I think maybe the conversation has now shifted.
00:19:16.240 --> 00:19:21.119
Well, it doesn't really matter, right, if if if you think AGI has happened or not.
00:19:21.279 --> 00:19:24.000
Like I said, it's a gray area, it's moving goldposts, et cetera.
00:19:24.240 --> 00:19:28.160
Because we are now, I think, focused on recursive self-improvement.
00:19:28.240 --> 00:19:43.119
And I think, uh, right, if if you look at Google's uh kind of four-tiered outlook um at going uh from you know, or toward superintelligence, one of the main paths there is uh recursive self-improvement going from AGI to ASI.
00:19:43.440 --> 00:19:45.039
So those are kind of the destinations.
00:19:45.119 --> 00:19:51.200
And uh I like to think of as as RSI as it's it's not a um a roadmap, right?
00:19:51.359 --> 00:19:57.119
Or it's not a a destination or a stopping point on that AI, AGI, ASI roadmap.
00:19:57.200 --> 00:20:00.720
It's more of the engine that can really uh expedite the process.
00:20:00.880 --> 00:20:06.799
And I think that one of the reasons why, if you go back and look, and I'm a dork, I did a show on this like two years ago.
00:20:06.880 --> 00:20:19.519
I went back and looked in archive.org, looked, you know, at every single main definition of artificial general intelligence and artificial superintelligence, you know, from 2005, 2010, 2015, 2020, right?
00:20:19.599 --> 00:20:24.319
And by those definitions, especially up through 2015, we're well past AGI.
00:20:24.480 --> 00:20:31.200
Um, but you know, RSI was never really a huge player in many of those earlier death definitions.
00:20:31.359 --> 00:20:35.279
And I think that that's why the timeline has kind of been expedited, right?
00:20:35.359 --> 00:20:42.720
A lot of people originally were like, oh, you know, AGI is a 20, you know, 2050, it's it's a 20, you know, 60, right?
00:20:42.799 --> 00:20:53.119
A lot of the earlier projections were saying, hey, we wouldn't have a single super smart AI system that could do work better than humans, you know, for 20, 30, 40 years.
00:20:53.200 --> 00:21:04.960
And that's clearly not the case because we are already there uh by you know um third-party metrics that measure those types of things, such as different GDP valve uh benchmarks.
00:21:05.279 --> 00:21:15.839
So the engine, though, that RSI engine, that's also why OpenAI's CEO, Sam Altman, just made a claim that I thought would actually get a lot more headlines, but it really didn't.
00:21:16.000 --> 00:21:26.000
So on a podcast last month, well, it's last month, but technically like a week and a half ago, he said that, well, we are now like in the singularity.
00:21:26.240 --> 00:21:26.640
All right.
00:21:26.880 --> 00:21:35.839
And usually when a CEO of one of the biggest companies in the world says something like that, um, it causes a little bit more attention.
00:21:35.920 --> 00:21:40.880
And maybe it didn't because of all the different things that have been happening uh over the past couple of weeks, right?
00:21:40.960 --> 00:21:48.000
We've seen these uh agents, uh, both from OpenAI and Anthropic uh kind of escaping their sandbox containments.
00:21:48.160 --> 00:21:53.359
So maybe that's why this statement from Sam Altman didn't get the uh the attention that maybe it deserves.
00:21:53.440 --> 00:21:59.039
But what he means by that is well, um, you know, in general, the singularity, right?
00:21:59.359 --> 00:22:18.240
So the Singularity is kind of like a hypothetical future point, uh, when AI surpasses human intelligence, but also it begins to recursively improve itself, and then it triggers or could trigger an uncontrollable explosive growth in technological progress.
00:22:18.559 --> 00:22:25.839
So I I don't know, by all intent, right, by everything that's going on, you can make an argument, right?
00:22:25.920 --> 00:22:37.759
Like a lot of uh, you know, naysayers or people um you know that kind of rally against you know what Sam Altman or any of the big AI CEOs say, you know, a lot of people say, Oh, you know, they're just saying this for marketing.
00:22:37.920 --> 00:22:46.720
But I don't know when you look at what's been happening, like I've said, over the last eight weeks, I wouldn't tend to agree, right?
00:22:47.039 --> 00:22:48.799
Are we in this singularity?
00:22:48.960 --> 00:23:02.079
Maybe, or at least maybe this is the uh the beginning, and we're getting into that phase where yes, we are at the point now uh where AI systems are well way smarter than humans can even comprehend.
00:23:02.240 --> 00:23:05.200
Um, and they're beginning to improve themselves.
00:23:05.359 --> 00:23:06.960
And when that happens, right?
00:23:07.119 --> 00:23:20.160
Um, kind of once RSI is full RSI, no human, that's when the pace of acceleration um gets so fast that even the people building it, you know, can't even understand or keep up.
00:23:20.319 --> 00:23:26.559
And I do think that, you know, maybe we are starting to enter uh that era of being in the singularity, right?
00:23:26.720 --> 00:23:28.240
Are we definitively there?
00:23:28.480 --> 00:23:29.200
Maybe not.
00:23:29.440 --> 00:23:31.680
Are we entering into that frame?
00:23:31.839 --> 00:23:33.039
I would say so, right?
00:23:33.119 --> 00:23:39.519
Like I've been the firm believer we've been, you know, well past the you know artificial general intelligence um era.
00:23:39.759 --> 00:23:45.599
Um, and we are probably more headed in this, you know, kind of the in-between uh phase, right?
00:23:45.680 --> 00:23:59.039
Which is this singularity, which is, you know, RSI, uh, you know, recursive self-improvement, leading to the point where, well, we are gonna see super intelligence probably, I don't know, in the next uh decade uh or less, right?
00:23:59.119 --> 00:24:00.880
I'd say that's probably a safe bet.
00:24:01.279 --> 00:24:04.640
So um it's it's a big claim that he said, right, on the podcast.
00:24:04.880 --> 00:24:18.079
Uh, but what he kind of meant by it is, you know, seeing this steady compounding progress where AI helps build the next AI, not necessarily the you know, Terminator Skynet scenario, but you know, here's kind of what exists today.
00:24:18.319 --> 00:24:23.680
So we don't have the full human-free RSI, right?
00:24:23.759 --> 00:24:24.960
And I want to make that clear.
00:24:25.039 --> 00:24:32.960
Uh, but because you're gonna be seeing a lot more about recursive self-improvement in the news as Anthropoc and Open AI go public, um, right?
00:24:33.039 --> 00:24:38.079
I'm sure they're gonna be timing their model releases around certain uh dates in terms of them going public.
00:24:38.160 --> 00:24:42.240
So there's gonna be a lot of confusion about, you know, what does all of this mean, right?
00:24:42.400 --> 00:24:44.799
What are the actual capabilities of these systems?
00:24:44.880 --> 00:24:49.359
But full RSI is where AI builds its successor, right?
00:24:49.519 --> 00:24:52.480
The next model, um, but it doesn't need humans.
00:24:52.559 --> 00:24:55.839
So maybe humans aren't overseeing it, um, right?
00:24:56.000 --> 00:25:05.440
So that piece hasn't happened yet, but we do know um that the big labs are using their current best models to help build, right?
00:25:05.519 --> 00:25:11.759
But it's probably a little bit more uh human oversight uh right now, or a little bit more human-led than full AI led.
00:25:11.839 --> 00:25:15.119
But Anthropic did say that that situation isn't inevitable.
00:25:15.279 --> 00:25:16.240
So what's real today?
00:25:16.400 --> 00:25:20.880
Well, the AI does the improvement work while humans still pick the goals and improve the results.
00:25:21.200 --> 00:25:27.039
But the AI is being able to handle exponentially more powerful and longer task jobs.
00:25:27.200 --> 00:25:34.880
So um, according to meter, um, AI, the task length is doubling roughly every four months, right?
00:25:35.599 --> 00:25:49.759
Which, I mean, the projections um, you know, by like next year at this time, you know, that a single AI model is going to be able to do uh work that would normally take a human like multiple weeks, right?
00:25:49.920 --> 00:25:57.359
Which is crazy to think about at a uh, I forgot if that's the 50% or the 80% pass rate, but regardless, right?
00:25:57.519 --> 00:26:01.680
Whereas, you know, two years ago, we weren't even having this conversation.
00:26:01.839 --> 00:26:10.880
We weren't even thinking of of you know the point where a model could work for you know hours or days, or we've even seen sometimes more than a week, right?
00:26:10.960 --> 00:26:22.400
The standard out-of-the-box models can work on these hard problems agentically, pulling tools, uh you know, starting to, you know, uh investigate one path, rewind, and go down another.
00:26:22.559 --> 00:26:24.400
Um, so now think of what happens.
00:26:24.480 --> 00:26:30.559
Well, when the labs are using this, and right, we what you and I use, right?
00:26:30.640 --> 00:26:34.319
The the GPT-5-6 souls and the fable fives, right?
00:26:34.480 --> 00:26:36.319
These these are not the best models, right?
00:26:36.400 --> 00:26:38.240
Obviously, all the companies they have better models.
00:26:38.319 --> 00:26:48.720
So think of when the better models that they have are actually building and improving those own versions that we aren't having yet, right?
00:26:48.799 --> 00:27:12.960
And and my hope is that ultimately what this means is well, maybe this this token maxing uh era was just a whiplash of waiting for uh the good some of the good benefits, some of the good early benefits of recursive self-improvement, uh, right, which is maybe just cheaper and more affordable AI, uh, at least if you are with a company that has properly invested in compute.
00:27:13.359 --> 00:27:16.799
So let's go back to um the Google Deep Mind.
00:27:16.880 --> 00:27:28.000
I know we've touched on this a couple of times, but uh this was Google Deep Mind's uh Jazjeet Shakan, who said that RSI is now central to the AI industry's investment thesis, uh, right?
00:27:28.079 --> 00:27:32.960
So he said this at a conference, and he was comparing that build out to the Apollo program as well.
00:27:33.039 --> 00:27:42.559
Um, and even talking about, you know, with that, Alphabet is planning to spend up to$205 billion um this year in CapEx, right?
00:27:42.640 --> 00:27:52.079
So if you don't know CapEx, that's capital expenditures, that's essentially, you know, the the hardware uh side, the data centers, the you know, the chips, the cooling, all that.
00:27:52.160 --> 00:28:10.799
Um, but the twist is this it's the same uh companies or the employees from those companies that are investing um into you know these hundreds of billions of dollars into ultimately the data infrastructure that makes recursive self-improvement possible.
00:28:10.960 --> 00:28:17.279
It is the employees from those companies that are actually saying, wait, this whole RSI thing, we might want to pace this out.
00:28:17.440 --> 00:28:17.759
All right.
00:28:17.839 --> 00:28:23.759
So a little bit more on this letter that was called Pacing the Frontier that came out last week.
00:28:23.920 --> 00:28:27.039
So the signers are some of the biggest names in AI.
00:28:27.119 --> 00:28:36.240
I mean, they include Anthropic CEO, um, open AI's chief scientist, um, and more than 1,300 verified employees from the big labs.
00:28:36.319 --> 00:28:40.480
A lot of people from uh OpenAI and Anthropic make up the biggest pack.
00:28:40.640 --> 00:28:47.200
You get uh some people from Google, uh, Meta, DeepMind, Thinking Machines, everyone, right?
00:28:47.359 --> 00:28:50.799
And they aren't asking technically for a pause.
00:28:50.960 --> 00:28:58.000
Some of them kind of are in their individual comments, but the letter themselves, or sorry, the letter itself is not asking for a pause.
00:28:58.160 --> 00:29:11.359
It is asking for the US government to essentially um start to make a plan, uh, both um domestically and eventually internationally, to say, well, what happens and what are our plans?
00:29:11.519 --> 00:29:18.880
Um, you know, once the AI development is too fast for us, the people building it, to keep up with, right?
00:29:19.200 --> 00:29:30.000
And I do think this is probably um one of the, and you do have to tip your hat to uh, you know, these researchers and also the companies themselves, right?
00:29:30.079 --> 00:29:37.200
A lot of them put out statements saying, hey, we support, you know, we support this letter or we support, you know, our employees who who sign this letter.
00:29:37.359 --> 00:29:42.160
Um, you know, some people look at this as you know being anti-accelerating, right?
00:29:42.240 --> 00:29:47.119
You know, there's been this, you know, this huge uh AI acceleration movement over the past few years, right?
00:29:47.279 --> 00:29:49.359
Accelerate at all costs, uh, right.
00:29:49.440 --> 00:29:55.200
You know, this can help us, you know, it's you know, get to this utopian, potential utopian future, right?
00:29:55.279 --> 00:29:58.079
But then there's the potential dystopian future as well, right?
00:29:58.240 --> 00:30:10.880
Each coin has two sides to it, and you can't just be focused on the utopian side uh without acknowledging the fact, well, there's a very dystopian side of what could happen if RSI or AGI ASI goes incredibly wrong.
00:30:11.039 --> 00:30:11.359
All right.
00:30:11.440 --> 00:30:18.000
So these in this letter, they're not asking the government to stop AI, they just want the tools to be ready to potentially slow down later.
00:30:18.160 --> 00:30:20.960
Um, but is it possible?
00:30:21.119 --> 00:30:22.160
Probably not, right?
00:30:22.319 --> 00:30:25.920
Um, I talked about this um on yesterday's show.
00:30:26.079 --> 00:30:30.559
I don't think you know China is going to be like, yes, we're gonna sign this, right?
00:30:30.720 --> 00:30:41.839
Ultimately, um, you know, even though China is still probably two months behind uh with their more open source open weight models approach, they're they have the next models ready as well, right?
00:30:41.920 --> 00:30:54.160
We've seen impressive releases from Kimi K38, or sorry, uh Kimi, uh Kimi K3, uh Quen 3.8, uh GLM5.2, you know, reportedly GLM53 is on the way, right?
00:30:54.559 --> 00:31:04.720
I get I get the need, um, and and I very much so respect uh, you know, because it takes a lot of guts to put your sign your name on a paper like this, um, on a letter like this.
00:31:04.960 --> 00:31:12.480
But at the same time, I I give China a very low likelihood of um actually slowing down their pace.
00:31:12.640 --> 00:31:32.960
So it is good to have the foundations in place when and if things may go awry, and then the very rare case that the US and China on something as important as artificial intelligence, which I think is gonna become more important uh, you know, ultimately than any natural resources, gold, oil, uh, it's gonna become more important than you know, companies military.
00:31:33.039 --> 00:31:39.279
Uh, you know, controlling AI ultimately means you take the driver's seat for being the global superpower.
00:31:39.359 --> 00:31:42.240
So I don't think China's gonna slow down, anyways.
00:31:42.319 --> 00:31:48.000
Um, I think the biggest worry though from those signing the letters is well, who's gonna ultimately check AI's work?
00:31:48.160 --> 00:31:51.839
Uh, a couple quotes that I thought were relevant from that.
00:31:52.000 --> 00:31:54.240
Um, so I wanted to read three of these.
00:31:54.400 --> 00:31:55.359
I think I have three of them up.
00:31:55.519 --> 00:31:55.920
Yeah, I do.
00:31:56.000 --> 00:31:57.039
Oh no, just two.
00:31:57.359 --> 00:32:12.319
Okay, so one was from uh hopefully these names right, uh Elena Slocum, a member of technical staff at Anthropic, who said automated AI research is a technology that will profoundly alter the course of human history, for better or for worse.
00:32:12.480 --> 00:32:18.640
It is imperative that we coordinate our efforts to ensure this technology becomes a boon for all mankind.
00:32:18.799 --> 00:32:29.119
Uh, and then uh Sheng Jai Zhao, the chief scientist at Meta, said this AI is progressing at a rate that our society might not be ready for.
00:32:29.279 --> 00:32:36.480
Frontier labs are very close to an AI that can exceed even the best people on almost every metric of intelligence.
00:32:36.640 --> 00:32:40.720
This will lead to unprecedented social and safety risks.
00:32:40.880 --> 00:32:46.160
To ensure a positive future, we need to develop AI in a way that is driven by responsibility and thoughtfulness.
00:32:46.480 --> 00:32:51.200
So, um here's here's my my blunt takeaway from this.
00:32:52.000 --> 00:33:00.160
Obviously, these researchers, scientists know far more about the capabilities than we do today, right?
00:33:00.240 --> 00:33:06.640
We're we're essentially living in archaic times already, even with today's best publicly available models.
00:33:06.799 --> 00:33:20.000
Because not only, right, I've talked about it, we're probably gonna see uh uh GPT-5.7 or a GPT-6, you know, next month, we're probably gonna be seeing a Fable 5.1 fairly soon, you know, maybe even in August.
00:33:20.160 --> 00:33:20.880
We'll see.
00:33:21.039 --> 00:33:21.759
Uh right.
00:33:21.920 --> 00:33:36.079
And and these companies are probably already working on the next version that comes after this, because my assumption is um a lot of those models are you know post-training and there's probably already you know new runs starting on the next models, anyways.
00:33:36.319 --> 00:33:41.359
They know what's coming next because they're working on it, they're researching it, right?
00:33:41.519 --> 00:33:48.559
So I I think a lot of people, again, are saying, well, you know, hey, I'm just using Copilot to improve my emails.
00:33:48.799 --> 00:33:50.480
Like, what are these people talking about?
00:33:50.720 --> 00:33:51.759
They know, right?
00:33:51.920 --> 00:34:08.719
Even if you are in the bubble a little bit more and you've been able to uh kind of see how powerful today's systems are, uh, right, being able to automate a large part of your old job description, um, what's coming next seemingly is even more powerful and potentially more worrisome.
00:34:08.880 --> 00:34:09.840
So here's the catch.
00:34:09.920 --> 00:34:13.599
I think that self-improving AI needs a trustworthy scorekeeper.
00:34:13.840 --> 00:34:28.079
And I think that's the crux of what this um pacing the frontier uh letter was ultimately getting at, you know, that a self-improving AI that gets better, you know, at whatever it's scoring system rewards, even if it's wrong, right?
00:34:28.159 --> 00:34:31.840
And that's where we're kind of seeing uh these containment breaks, right?
00:34:31.920 --> 00:34:33.679
Because these AI models, right?
00:34:33.760 --> 00:34:39.920
If if you say, hey, you're gonna be rewarded for doing the best on this benchmark, and then it breaks the containment, right?
00:34:40.000 --> 00:34:43.119
And it's saying, well, I'm just doing my job, right?
00:34:43.199 --> 00:35:01.519
So, you know, you need a better way that, you know, especially as we enter a more full era uh of recursive self-improvement to um, you know, reward these models and to make sure that there's uh more trustworthy systems in place, you know, during the training and post-training processes.
00:35:01.760 --> 00:35:11.440
So we've seen multiple instances of this, like I've said, but I think it's so far been relatively harmless agent outbreak over the past 10 days, both from uh open AI and anthropic.
00:35:11.599 --> 00:35:16.239
But once RSI arrives, those kind of outbreaks could be catastrophic.
00:35:16.400 --> 00:35:20.000
Not, and I'm not saying from open AI and anthropic, that's not what I'm saying.
00:35:20.159 --> 00:35:21.920
I am saying ultimately, right?
00:35:22.000 --> 00:35:33.679
And I said this in yesterday's show, um, in in two years, right, on consumer hardware, you're gonna have models that are more capable than Fable 5.
00:35:33.760 --> 00:35:39.840
You're gonna have models that are more capable uh than GPT-5.6 soul on consumer desktops, right?
00:35:40.000 --> 00:35:40.960
Open source.
00:35:41.119 --> 00:35:43.920
Um, and that's where I start to worry about, right?
00:35:44.079 --> 00:35:59.760
When you have a version of an RSI model that you know you could be running on your computer, and that could be spitting out, you know, dozens or hundreds, who knows, of new smaller models a day that are super smart at super tasks, right?
00:35:59.840 --> 00:36:11.440
So that it's it's more of you know, when the guardrails could be let out, which is you know, open source does open that Pandora's box um of of you know how AI could be used in a bad way, right?
00:36:11.519 --> 00:36:21.039
At the same time, it does give defenders, I think, more access to more tools uh to be able to uh, you know, make the internet and hopefully the physical world a safer place.
00:36:21.199 --> 00:36:29.119
Um, but I think that humans are still gonna have to own that judgment role, even as RSI becomes more commonplace.
00:36:29.199 --> 00:36:33.519
And that's exactly where, as we wrap, I want your business strategy to start.
00:36:33.679 --> 00:36:36.480
So I threw a lot of information up uh at you.
00:36:36.559 --> 00:36:38.400
I wanted a couple tangents, right?
00:36:38.480 --> 00:36:41.679
That's why this thing's, you know, largely unedited, unscripted.
00:36:41.760 --> 00:36:47.039
I want you to get, you know, just kind of um some some real advice and and thoughts and feedback.
00:36:47.119 --> 00:36:49.840
But as we wrap up, what does all this mean for you?
00:36:50.159 --> 00:36:50.719
Right?
00:36:50.960 --> 00:37:07.360
Hey, if ultimately you're just trying to, you know, make your um, you know, agents more secure, uh, if you're just trying to increase productivity, if you're just trying to get more out of everyday, you know, you know, your um your AI systems that you have in place at your company.
00:37:07.519 --> 00:37:08.480
Why does this matter?
00:37:08.719 --> 00:37:09.679
Why does this matter?
00:37:09.840 --> 00:37:14.079
Well, you have to design for the loop, I think, or you're gonna get dragged.
00:37:14.239 --> 00:37:15.360
Here's what I mean by that.
00:37:15.519 --> 00:37:22.719
You have to plan, I think, for AI prices to actually fall, which you might have just gotten wind on this.
00:37:22.800 --> 00:37:25.760
Oh my gosh, we need to cut back on token spent.
00:37:25.920 --> 00:37:44.480
I think we have to look past that because the labs that have properly dedicated enough compute um to maybe achieve um RSI or, you know, uh what some labs are calling an automated AI researcher, those are the companies that, well, they're gonna be able to make prices cheaper.
00:37:44.559 --> 00:37:54.320
And we quite literally saw that um with OpenAI uh decreasing the price of GPT-5.6 Luna, a very capable model, right?
00:37:54.400 --> 00:37:57.199
You can't just say, oh, it's their smallest model, it's not any good.
00:37:57.280 --> 00:37:58.480
No, it's very good.
00:37:58.639 --> 00:38:05.119
Um, right, it is much better than you know, GPT-5, GPD 5.1, 2, 3, 4, right?
00:38:05.199 --> 00:38:09.920
It is a very capable model, and it is like free 99 right now.
00:38:10.079 --> 00:38:22.719
Um, so I think you have to prepare for prices to ultimately fall, but I think you ultimately have to build modularly because maybe you know you're with a different provider, and well, maybe your provider, right?
00:38:22.800 --> 00:38:30.159
At least when we look at the big four, the big six, maybe they're gonna figure out RSI a year after the whoever figures it out first.
00:38:30.320 --> 00:38:34.480
So that's why I think you can't be too entrenched uh with one provider.
00:38:34.639 --> 00:38:38.719
Um, I think you should be, you know, always choose your AI operating system of choice.
00:38:38.800 --> 00:38:39.679
I'm very transparent.
00:38:39.760 --> 00:38:45.039
I've always said for me and for most companies, I advise, I say that's Chad GPT because it's the easiest to learn, right?
00:38:45.119 --> 00:38:50.000
But whether you're anthropic, co-pilot, Gemini, Chad GPT, uh, Grok, Meta, right?
00:38:50.079 --> 00:38:56.639
I don't know anyone that any companies that are necessarily have chosen that as their AI operating system, but you should always be looking at fallbacks, right?
00:38:56.719 --> 00:39:01.119
Whether that's another frontier proprietary uh provider, whether that's an open model.
00:39:01.280 --> 00:39:04.079
Um, but you should also be copying the labs, right?
00:39:04.239 --> 00:39:08.639
They're using their expensive models for the hardest tasks only, right?
00:39:08.719 --> 00:39:11.760
And then they're using uh, you know, other models, right?
00:39:11.840 --> 00:39:16.719
So they're using that for judgment and planning, but then they're using cheaper models for the busy work, and you should be doing that too.
00:39:16.960 --> 00:39:17.760
I've already said this.
00:39:17.840 --> 00:39:31.280
We're at the point now with the capability overhangs where you don't need a Fable Five and a GPT 5.6 soul to rewrite your emails or to do the equivalent of a tough Google search, right?
00:39:31.360 --> 00:39:34.559
Or a simple uh you know, looking up some facts.
00:39:34.639 --> 00:39:39.920
Um, you know, you have to have um that kind of separated.
00:39:40.079 --> 00:39:42.880
Um, also another way to start copying the labs.
00:39:42.960 --> 00:39:45.039
This is something I've been doing now for a couple of months.
00:39:45.280 --> 00:39:52.880
Well, is you should have your AIs control other AIs and you should uh right, you maybe start building other AIs, right?
00:39:52.960 --> 00:39:57.840
And I think we didn't really get into that uh that realm until we got to GPT 5.6 soul, right?
00:39:57.920 --> 00:40:08.400
You had some people share online, and I talked about in the show about how people with no background in creating models were able to create, you know, small models using GBT5.
00:40:12.559 --> 00:40:19.360
Is someone used uh GPT 5.6 soul to look in, look at every single text message that they had ever sent via iMessage, right?
00:40:19.440 --> 00:40:31.360
So they had soul work on this and they created a literal small language model with weights um that was just based on here's you know everything I text about, here's how I write, here's and you know, and it knows everything about me, right?
00:40:31.440 --> 00:40:32.800
So, so think of that, right?
00:40:32.960 --> 00:40:38.880
Think of how when when prices I think are ultimately going to be driven down as the models get more powerful.
00:40:39.039 --> 00:40:41.519
What are those moves that you can make and copy from the big labs?
00:40:41.599 --> 00:40:46.880
And then last but not least, the question isn't whether about AI builds AI.
00:40:47.039 --> 00:40:51.679
The question isn't about, you know, oh my gosh, what does that mean, you know, recursive self-improvement?
00:40:51.840 --> 00:40:53.199
You know, what does that mean for our business?
00:40:53.360 --> 00:40:54.559
Oh my gosh, freak out.
00:40:54.719 --> 00:40:57.280
No, it's just whether your business is ready.
00:40:57.519 --> 00:41:02.559
So you have to start thinking and talking openly about the implications, right?
00:41:02.719 --> 00:41:15.440
What happens, you know, when there's maybe dozens of AI models that are um distinct and made for your job, for your department, for your sector, because that time is coming.
00:41:15.599 --> 00:41:30.320
Because I think ultimately what recursive self-improvement means, aside from hopefully uh, well, better models coming out faster, but also cheaper prices and probably more specific in niche uh models that serve specific verticals.
00:41:30.400 --> 00:41:38.320
So you have to start replanning now in whatever systems that you've got in place in 2024, 2025, etc., those can't be long-standing systems.
00:41:38.400 --> 00:41:42.880
I talked, you can't be planning a year at a time, you should be planning a month at a time.
00:41:42.960 --> 00:41:48.400
And I know that's you know, uh kind of daunting for traditional digital transformation, but I think that's where you have to be at.
00:41:48.559 --> 00:41:48.880
All right.
00:41:48.960 --> 00:42:04.480
I hope this show was helpful going over a lot of these terms, RSI, but hopefully now you know a little bit more about recursive self-improvement, what it means when these models are kind of improving themselves or working on the next version of themselves or making AI maybe cheaper and faster and better for all of us.
00:42:04.639 --> 00:42:11.599
So now you know what happens when RSI and models start improving themselves and hopefully what it means for your business.
00:42:11.840 --> 00:42:25.039
If this was helpful, please consider reposting this if you're listening on social media, if you're listening on the LinkedIn machine, uh, and if you're listening to the podcast on Spotify or Apple Music, please subscribe, then go to your everydayai.com.
00:42:25.199 --> 00:42:26.079
So thank you for tuning in.
00:42:26.159 --> 00:42:29.039
Hope to see you back tomorrow and every day for more everyday AI.
00:42:29.119 --> 00:42:30.159
Thanks, y'all.
00:00:02.479 --> 00:00:20.879
As Agentic AI became more powerful and token hungry in the first months of 2026, the AI decision makers a month or so ago started to collectively tighten their token maxing belts to prepare for a more expensive AI future.
00:00:21.039 --> 00:00:21.359
Right?
00:00:21.519 --> 00:00:26.559
It's the shift from token maxing to token efficiency we've been talking about here for months.
00:00:26.800 --> 00:00:33.520
But then something drastic and kind of unexpected happened that I don't think enough people are talking about.
00:00:33.759 --> 00:00:50.159
OpenAI on Thursday revealed that its GPT-5.6 soul model improved itself and actually improved itself so much after its initial release that they were cutting prices on some of the variants of the model by 80%.
00:00:51.039 --> 00:00:53.200
So we've reached kind of a new juncture.
00:00:53.359 --> 00:01:02.079
Today's Frontier models are not just helping to create or distill their smaller variants, but they're actually improving themselves after release.
00:01:02.320 --> 00:01:09.519
Hence the next AI buzzword you'll be hearing a lot over the next year: recursive self-improvement or RSI.
00:01:09.840 --> 00:01:20.159
So, in short, recursive self-improvement is a process where an artificial intelligence system uses its own capabilities to design, code, and build a more advanced version or a better version of itself.
00:01:20.319 --> 00:01:33.439
And you'll be hearing this a lot more recently because RSI has become just the dominant tech headline over the past few weeks with a handful of events that we're gonna break down on today's show.
00:01:33.599 --> 00:01:48.959
But unlike so many buzzwords that we've covered here on everyday AI over the years, RSI is one that actually today's AI implementations are gonna be impacted by, and so too will tomorrow's AI security.
00:01:49.200 --> 00:02:02.400
In practical sense, like in OpenAI's case, RSI can change what AI capabilities companies everywhere can actually afford, but it also sheds light on potential downsides of self-improving models, doubling down the wrong path.
00:02:02.640 --> 00:02:08.719
So, what the heck is RSI and why is every recent AI story touching on it?
00:02:08.800 --> 00:02:10.319
And what do we all need to know?
00:02:10.479 --> 00:02:13.039
Well, let's jump straight into the big picture.
00:02:13.280 --> 00:02:18.400
The big picture here is AI has started building better and more efficient AI.
00:02:18.560 --> 00:02:24.800
So Frontier AI models now improve themselves, they're rewriting their own code and training their smaller siblings.
00:02:24.960 --> 00:02:34.159
So this is different than when you know Anthropic and OpenAI are talking about using their own models to build products like Claude Code or like Codex.
00:02:34.479 --> 00:02:46.159
This is when the models are actually making themselves better, and that process is called RSI, and it's no longer some future science fiction, it is a near-term reality.
00:02:46.319 --> 00:02:56.719
So OpenAI, as an example, says that its GPT-5.6 soul model made its Luna model, the smaller variant, just cheaper and more efficient.
00:02:56.879 --> 00:02:58.719
And then they cut prices by 80%.
00:02:59.520 --> 00:03:05.039
But the full human-free recursive self-improvement hasn't yet arrived.
00:03:05.280 --> 00:03:19.680
But over the past two months, there's a lot of AI news stories that we're gonna be connecting the dots on uh the dots on today that show that, well, this probably isn't like a 2029-2028 thing that a lot of people have been saying.
00:03:19.840 --> 00:03:26.159
It's probably a 2027 thing, and that changes things maybe in both a good and a scary way.
00:03:26.560 --> 00:03:28.639
So, on today's show, here's where you're gonna learn.
00:03:28.800 --> 00:03:37.120
You're gonna learn why OpenAI was able to cut prices up to 80% when everyone predicted that AI costs would continue to climb in 2026.
00:03:37.199 --> 00:03:44.479
You're gonna know more about this eight eight-week timeline that turned RSI from research jargon into mainstream headlines.
00:03:44.639 --> 00:03:55.520
You're gonna understand by the end of today's show what RSI actually means in plain English and what Sam Altman, the CEO of OpenAI's singularity claim that he recently made actually means.
00:03:55.680 --> 00:04:06.639
And you're gonna know why the people building RSI are actually asking for maybe the industry to pause a little bit and pace itself and how your business should respond to all of this.
00:04:06.879 --> 00:04:07.919
All right, let's get into it.
00:04:08.000 --> 00:04:09.199
Welcome to Everyday AI.
00:04:09.360 --> 00:04:11.759
My name is Jordan Malton, and well, this thing's for you.
00:04:11.919 --> 00:04:18.319
This is your daily unedited, unscripted live stream podcast, and free daily newsletter helping business leaders like you and me keep up.
00:04:18.399 --> 00:04:19.839
I do all the hard research.
00:04:20.000 --> 00:04:25.600
So you sit back, listen, read our newsletter, enjoy the benefits, and you grow your company and career.
00:04:25.759 --> 00:04:31.279
So it starts here, but make sure to subscribe to our free daily newsletter at your everydayai.com.
00:04:31.360 --> 00:04:36.560
We're going to be recapping all the highlights from today's show as well as all of the other AI news that you need to know.
00:04:36.879 --> 00:04:46.160
So let's get into the latest uh three to four letter acronym in our ever-evolving bowl of alphabet AI soup.
00:04:46.240 --> 00:04:47.759
We are talking RSI.
00:04:48.720 --> 00:04:53.199
So let's talk about the last uh two months.
00:04:53.519 --> 00:05:03.439
Because I mean, yes, I've been talking about uh recursive self-improvement on this show probably for the last two or so years, since I've been doing this for the last three and a half years.
00:05:03.600 --> 00:05:09.600
Um, as it seemed like recursive self-improvement was actually going to be something that happened this decade.
00:05:09.680 --> 00:05:10.319
Um, right.
00:05:10.399 --> 00:05:18.399
Because if you when I started this thing in 2023, most people thought that recursive self-improvement was at best a 2030 thing.
00:05:18.560 --> 00:05:32.000
Yet here we are, where we're already getting glimpses and sniffs of real RSI, and probably we will get full RSI by next year, which is actually crazy to think about.
00:05:32.160 --> 00:05:38.639
So, but let's kind of uh look at the last two months and some of the different uh dots that we're gonna connect here.
00:05:38.720 --> 00:05:44.959
So in June, Anthropic published a post on its website called When AI Builds Itself.
00:05:45.120 --> 00:05:55.120
Uh, also in June, Google researchers mapped four different routes from AGI to superintelligence, and one of those routes well relied heavily on recursive self-improvement.
00:05:55.279 --> 00:06:04.800
Uh, then in July, uh OpenAI launched an internal RSI benchmark, uh, measuring different models' ability to well improve themselves.
00:06:04.959 --> 00:06:11.759
Then, like I said, its soul model post-trained its smaller sibling Luna at the end of July.
00:06:11.920 --> 00:06:18.639
And then also in a podcast interview, uh, OpenAI CEO Sam Altman declared that we are in the singularity.
00:06:18.720 --> 00:06:34.639
Uh yeah, and the news cycle was just warming up because uh at the end of July, so last week, startup recursive superintelligence signed a$400 million AWS compute deal to automate AI research.
00:06:34.720 --> 00:06:38.879
So not only is it the labs that all of us know, right?
00:06:38.959 --> 00:06:50.879
Kind of like the big four plus meta and acts, all working toward, you know, some version of recursive self-improvement, but it's also, you know, new startups that are coming in heavily funded just to do that and nothing else.
00:06:51.040 --> 00:07:26.399
And on that same day, so on July 28th, over a thousand, and I think that number is like 1300 now, Frontier Lab employees published a paper called Pacing the Frontier, which essentially is kind of asking the governments, uh, the US governments to maybe slow down or at least prepare for the possibility that there needs to be some international um kind of agreement when it comes to the pace of AI, because a lot of people signing that letter in the comments said that, well, AI is developing maybe too quickly and that society can't keep up.
00:07:26.639 --> 00:07:33.519
And let me just say this right now, um, and I need to hit a pause because I think that, you know, I am not in the Silicon Valley bubble, right?
00:07:33.600 --> 00:07:38.000
I don't work at a big tech company, um, but I talk about AI every single day.
00:07:38.079 --> 00:07:43.120
And I do talk to those people building uh AI very recently uh very frequently.
00:07:43.360 --> 00:07:48.079
And if you're listening to the show, you're probably uh, you know, more like me, right?
00:07:48.240 --> 00:07:53.600
I know we have a lot of people um at those big tech companies, but um I'll say this.
00:07:54.560 --> 00:08:02.959
The rest of the world right now, when they're viewing artificial intelligence, right, most people are saying, oh yeah, I use Copilot to make my emails better.
00:08:03.120 --> 00:08:08.720
Or they're saying, you know, oh, I use, you know, uh Chat GPT now instead of Google, uh, right.
00:08:08.879 --> 00:08:14.720
Or they're saying, oh, I use, you know, Google Gemini to, you know, make, you know, cool graphics, right?
00:08:15.120 --> 00:08:16.000
That's it.
00:08:16.240 --> 00:08:20.240
Um, so I would say that, you know, aside from our audience, right?
00:08:20.399 --> 00:08:24.000
I want you to think of yourself right now as you're kind of in that AI bubble, right?
00:08:24.079 --> 00:08:28.399
Especially if you're a frequent listener or if you've been listening for a long time.
00:08:28.560 --> 00:08:32.000
You know, we're kind of in that bubble, even though we're not in Silicon Valley, right?
00:08:32.080 --> 00:08:37.840
We understand how uh capable these models are, but the rest of the world really has no clue, right?
00:08:38.000 --> 00:08:46.720
All they're seeing is they're seeing, you know, oh, you know, all these AI models, you know, I'm just going in here, I'm finding the one I want, and you know, maybe my work's a little bit easier.
00:08:46.879 --> 00:08:53.279
Or, you know, I think that there's a lot of people now looking at AI, and there is this uh this recent anti-AI swing, right?
00:08:53.360 --> 00:09:00.080
When it comes to data centers and, you know, the college graduates that were booing, you know, commencement speakers who are talking about AI.
00:09:00.159 --> 00:09:06.159
So there's a lot of just bad information uh out there when it comes to AI that has created this anti-AI backlash.
00:09:06.320 --> 00:09:16.960
So I think, if nothing else, there is such a huge disconnect between the general uh, you know, non everyday AI listener type, right?
00:09:17.120 --> 00:09:23.600
Um, or you know, think of your in your organization, you know, there's probably a team of AI champions and then there's everyone else, right?
00:09:23.759 --> 00:09:34.799
So for I would say 90% of the US working population, uh, they really don't have any clue what today's AI systems are capable of, right?
00:09:34.960 --> 00:09:49.919
If you sat all those people down and showed them uh what something like, you know, GPT-5-6 soul or something like you know, Fable uh five and you know, in Codecs and Claude Code, if you show them what it could do, they would not believe you.
00:09:50.080 --> 00:09:51.759
They would say, oh, this is from the future, right?
00:09:51.840 --> 00:10:02.639
And I think that's why we got this letter from the Frontier Labs called Pacing the Frontier, uh, which we'll talk about here in a couple of minutes, that are saying, like, hey, we might need to pace uh the rate of AI acceleration.
00:10:02.799 --> 00:10:18.480
And then days later, uh, which is kind of a sounds like a small uh side plot, but it's not a formal OpenAI employee uh who became a thinking uh Thinking Labs co-founder rejoined OpenAI to specifically um work on RSI.
00:10:18.559 --> 00:10:22.399
And then Google also, right, all these things happening at once.
00:10:22.480 --> 00:10:25.679
Uh, you know, this was uh the end of July as well.
00:10:25.919 --> 00:10:30.559
Uh a Google exec uh called CapEx an RSI bet.
00:10:30.799 --> 00:10:31.840
So what does that mean, right?
00:10:31.919 --> 00:10:36.399
Uh collectively, the industry is spending hundreds of billions of dollars on data centers.
00:10:36.559 --> 00:10:41.360
And essentially a Google exec said, well, this is because of recursive self-improvement, right?
00:10:41.519 --> 00:10:44.159
You need all this compute to be able to do this, right?
00:10:44.240 --> 00:10:44.960
That's the difference.
00:10:45.120 --> 00:10:56.399
You know, when you have a team of researchers, uh, maybe, you know, three to four years ago, where this kind of the AI researcher wasn't yet automated or on its way to being automated, right?
00:10:56.559 --> 00:10:58.799
You just had to use a lot more people.
00:10:59.120 --> 00:10:59.919
Um, right.
00:11:00.080 --> 00:11:05.440
Yes, you were still able to, you know, the concept of uh recursive self-improvement is not new.
00:11:05.519 --> 00:11:14.639
It's been around since, you know, I would say like 2016, technically, with Alpha Go, um, you know, and also some of the earlier GPT models, uh, right.
00:11:14.720 --> 00:11:20.799
But that was much more kind of human hand holding, uh, you know, AI models to improve, right?
00:11:20.960 --> 00:11:26.080
Where now it's more of uh human oversight and the models are kind of doing a lot of the work on themselves.
00:11:26.159 --> 00:11:29.840
So, like I said, it's not full human-free, uh, but we are at the point now.
00:11:30.080 --> 00:11:44.080
But, you know, a high-ranking Google exec essentially just said last week that yeah, all this money that's being spent on these data centers, this this hundreds of billions of dollars, that's so that is for recursive self-improvement.
00:11:44.159 --> 00:11:56.559
So these models are just gonna be able to build not only the next version of themselves, but once that next version has been released, like OpenAI just did, well, it's gonna improve itself to hopefully make it cheaper, faster, better, right?
00:11:56.720 --> 00:12:05.519
But then also, I think, right, I've always been saying I'm a huge believer in this concept that eventually there's gonna be thousands of small models, uh, right.
00:12:05.679 --> 00:12:07.039
I think that that's gonna be the truth.
00:12:07.120 --> 00:12:12.320
There's gonna be thousands of models that help with marketing, there's gonna be thousands of models that help with uh medicine.
00:12:12.399 --> 00:12:14.960
There's right, there's gonna be thousands, but small ones.
00:12:15.039 --> 00:12:25.759
And these this is where I think the future is heading is recursive self-improvement, where uh these model makers are gonna be able to spin out, you know, probably the ones first that make the most sense to make money, right?
00:12:25.840 --> 00:12:30.000
And then from after that, I'm guessing we'll we're gonna equally see or hopefully see.
00:12:30.080 --> 00:12:37.840
It's my hope that we'll uh you know, equally see uh, you know, all of these small models that come from RSI that are gonna be able to do good in the world, right?
00:12:37.919 --> 00:12:44.000
And we're already seeing that uh with things like medicine and biology and you know, right now, uh defensive cybersecurity.
00:12:44.240 --> 00:12:51.840
So now let's talk about that Thursday shocker that kind of changed, I think, the near-term AI strategy for a lot of business owners out there.
00:12:52.000 --> 00:13:03.039
Uh so for months, right, we've been hearing this, this, this concept um that these powerful AI agents, well, they were too token hungry, you know.
00:13:03.200 --> 00:13:13.039
So I think in late 2025, kind of with the the advent of of Claude Code, Claude Cowork in early 2026, uh, and then codex in February, right?
00:13:13.120 --> 00:13:19.120
It kind of got to this point where everyone was spending as many tokens as possible because companies everywhere were like, oh my gosh, right?
00:13:19.200 --> 00:13:26.159
Those on the bleeding edge, those AI champions were like, my gosh, these systems can do absolutely anything, which was great, right?
00:13:26.240 --> 00:13:28.559
But the revenue didn't always directly follow.
00:13:28.720 --> 00:13:41.440
And so what that meant is while you know some companies were waiting to see the uh the fruits of their token maxing labor, they had to cut back, right, and focus a little bit more on token efficiency or as some people say, value maxing.
00:13:41.600 --> 00:13:47.759
So we kind of thought that at this point, well, okay, we're just gonna probably start spending less, right?
00:13:48.000 --> 00:13:59.759
But open AI came and threw that out the window because open AI said that Seoul, their GPD-5.6 soul, adapted an existing post-training setup for the smaller Luna model.
00:13:59.919 --> 00:14:04.000
So essentially, right, and they actually shared this, and we shared this in our newsletter as well.
00:14:04.159 --> 00:14:08.639
Uh, they kind of shared and they did this, I believe, all in codecs.
00:14:08.799 --> 00:14:14.559
Um, so it's it's it's you know, wasn't some far off, you know, science fiction.
00:14:14.639 --> 00:14:28.799
It's some uh uh according to someone that tweeted about this, um, it is some open AI researchers who used codecs and they used the GBT56 soul model to improve the smaller versions.
00:14:28.960 --> 00:14:31.440
Uh, so the medium size is GBT5.
00:14:35.759 --> 00:14:41.200
And then GBD56 Luna, which they were able to cut the prices on that by 80%, right?
00:14:41.440 --> 00:14:45.200
And the fact that that happened, and I still don't think enough people are talking about it, right?
00:14:45.360 --> 00:14:53.200
Because if you compare as an example, um, and I did talk about this on the show uh yesterday, maybe, or maybe it was Friday, um, right?
00:14:53.360 --> 00:15:06.879
But now you have GBD56 Luna, which according to artificial uh analysis is roughly about the same as uh Claude Sonnet 5, but it is 25x cheaper per task completion, right?
00:15:07.120 --> 00:15:15.440
So now it's almost like, wait, you know, there is this narrative in quarter two that we were gonna have to start using less AI.
00:15:15.600 --> 00:15:25.840
And now all of a sudden, because we are getting this hint of recursive self-improvement from open AI first, now we have to start rethinking that strategy immediately.
00:15:26.399 --> 00:15:28.240
So let's define it a little bit.
00:15:28.320 --> 00:15:28.960
Uh, all right.
00:15:29.120 --> 00:15:36.159
So uh RSI is essentially when an AI helps improve another AI, it could be a from a different company, right?
00:15:36.240 --> 00:15:40.080
It could be its own um smaller version, like we saw with the G V D5.
00:15:41.279 --> 00:15:43.600
Um uh working on G V D5.
00:15:44.720 --> 00:15:50.960
And we've seen from other companies talking about kind of their big model helping train some of their smaller models.
00:15:51.039 --> 00:15:54.320
Uh, and we've seen also this is similar to distillation, right?
00:15:54.480 --> 00:15:56.799
So um intercompany distillation.
00:15:56.960 --> 00:16:08.559
Um, you know, generally when they're talking about that, yes, there's humans involved, but I would say in 2026, uh, I I think the labs will start talking about this more as it becomes more commonplace.
00:16:08.720 --> 00:16:17.440
Um, but you know, now it's with these smaller models, it is very commonplace for companies to distill the smaller versions from the big versions.
00:16:17.600 --> 00:16:23.039
And I think that's also why we're sometimes seeing um the versions come out at a different time, right?
00:16:23.200 --> 00:16:26.559
So, as an example, Fable 5 came first, right?
00:16:26.720 --> 00:16:33.840
And then we or you know, Fable and Mythos 5 came first, and then we got you know Sonnet 5, and then we got Opus five.
00:16:33.919 --> 00:16:43.840
So uh, you know, Anthropic has kind of hinted in the past that a lot of their code, up to you know, 90%, I think according to some estimates, you know, 90% of their code is actually written by their own models.
00:16:44.399 --> 00:16:54.720
So uh, but the Google exec though, um, I think this is important, um, compared it to history, talking about steam engines were used to build the next steam engine.
00:16:54.879 --> 00:17:01.840
And I think to understand why that matters, uh, we have to take a look at the um all the other definitions, right?
00:17:01.919 --> 00:17:06.480
Um, and and how recursive self-improvement actually fits on the current timeline.
00:17:06.640 --> 00:17:22.240
Because for the most part, uh it it, at least if you've been following uh and keeping track at home, the um kind of the the path uh to um you know where AI is eventually going to go, whether you're excited about that or uh not.
00:17:22.480 --> 00:17:26.720
And you know, I constantly find myself going on on each side of that equation, right?
00:17:26.799 --> 00:17:35.599
But you think of, well, we're in the artificial intelligence um era, and it seems like most companies are focused on super intelligence.
00:17:35.839 --> 00:17:40.799
Okay, so superintelligence is when, well, all AI is smarter than all humans.
00:17:40.960 --> 00:17:43.839
Um, but AGI is kind of that stepping point.
00:17:44.000 --> 00:17:53.519
So uh, you know, if you had to say three things, it's AI to AGI, which is artificial general intelligence, um, and then you have superintelligence.
00:17:53.680 --> 00:18:03.680
And that's I think superintelligence, if you uh if superintelligence goes wrong, that's where you start getting into the Terminator and Skynet uh type comparisons, right?
00:18:03.759 --> 00:18:10.720
When superintelligence goes right, that's when, well, in theory, we could cure most diseases, right?
00:18:10.960 --> 00:18:14.559
Um, but there's those kind of steps belong uh along the way.
00:18:14.720 --> 00:18:26.240
Um, and I think that uh according to um a lot of people's timelines, um actually RSI is kind of what pushes from AGI to ASI.
00:18:26.480 --> 00:18:30.000
Uh and I know we're throwing a lot of um acronyms out there.
00:18:30.079 --> 00:18:34.799
So I'm in the firm belief that we've already hit artificial general intelligence, right?
00:18:34.960 --> 00:18:38.160
I know it's kind of a touchy topic, it's it's moving goalposts.
00:18:38.319 --> 00:18:41.440
Uh, luckily, you know, I've been saying this since uh last year.
00:18:41.519 --> 00:18:44.799
Luckily, uh NVIDIA CEO Jensen Wong uh said something similar.
00:18:44.880 --> 00:18:49.920
So I don't feel as crazy when you know super smart people start saying, yeah, we're probably past AGI, right?
00:18:50.000 --> 00:18:59.119
So artificial general intelligence is well when most AI systems can produce economically viable work at the same rate or higher than most humans.
00:18:59.200 --> 00:18:59.759
Uh, right.
00:18:59.839 --> 00:19:16.079
And I think as we've seen uh agentic AI be able to use computers, being able to browse the internet, being able to see and hear and listen and do all these things technically, you know, through a computer that an average human sitting in front of a computer could do, you know, I think maybe the conversation has now shifted.
00:19:16.240 --> 00:19:21.119
Well, it doesn't really matter, right, if if if you think AGI has happened or not.
00:19:21.279 --> 00:19:24.000
Like I said, it's a gray area, it's moving goldposts, et cetera.
00:19:24.240 --> 00:19:28.160
Because we are now, I think, focused on recursive self-improvement.
00:19:28.240 --> 00:19:43.119
And I think, uh, right, if if you look at Google's uh kind of four-tiered outlook um at going uh from you know, or toward superintelligence, one of the main paths there is uh recursive self-improvement going from AGI to ASI.
00:19:43.440 --> 00:19:45.039
So those are kind of the destinations.
00:19:45.119 --> 00:19:51.200
And uh I like to think of as as RSI as it's it's not a um a roadmap, right?
00:19:51.359 --> 00:19:57.119
Or it's not a a destination or a stopping point on that AI, AGI, ASI roadmap.
00:19:57.200 --> 00:20:00.720
It's more of the engine that can really uh expedite the process.
00:20:00.880 --> 00:20:06.799
And I think that one of the reasons why, if you go back and look, and I'm a dork, I did a show on this like two years ago.
00:20:06.880 --> 00:20:19.519
I went back and looked in archive.org, looked, you know, at every single main definition of artificial general intelligence and artificial superintelligence, you know, from 2005, 2010, 2015, 2020, right?
00:20:19.599 --> 00:20:24.319
And by those definitions, especially up through 2015, we're well past AGI.
00:20:24.480 --> 00:20:31.200
Um, but you know, RSI was never really a huge player in many of those earlier death definitions.
00:20:31.359 --> 00:20:35.279
And I think that that's why the timeline has kind of been expedited, right?
00:20:35.359 --> 00:20:42.720
A lot of people originally were like, oh, you know, AGI is a 20, you know, 2050, it's it's a 20, you know, 60, right?
00:20:42.799 --> 00:20:53.119
A lot of the earlier projections were saying, hey, we wouldn't have a single super smart AI system that could do work better than humans, you know, for 20, 30, 40 years.
00:20:53.200 --> 00:21:04.960
And that's clearly not the case because we are already there uh by you know um third-party metrics that measure those types of things, such as different GDP valve uh benchmarks.
00:21:05.279 --> 00:21:15.839
So the engine, though, that RSI engine, that's also why OpenAI's CEO, Sam Altman, just made a claim that I thought would actually get a lot more headlines, but it really didn't.
00:21:16.000 --> 00:21:26.000
So on a podcast last month, well, it's last month, but technically like a week and a half ago, he said that, well, we are now like in the singularity.
00:21:26.240 --> 00:21:26.640
All right.
00:21:26.880 --> 00:21:35.839
And usually when a CEO of one of the biggest companies in the world says something like that, um, it causes a little bit more attention.
00:21:35.920 --> 00:21:40.880
And maybe it didn't because of all the different things that have been happening uh over the past couple of weeks, right?
00:21:40.960 --> 00:21:48.000
We've seen these uh agents, uh, both from OpenAI and Anthropic uh kind of escaping their sandbox containments.
00:21:48.160 --> 00:21:53.359
So maybe that's why this statement from Sam Altman didn't get the uh the attention that maybe it deserves.
00:21:53.440 --> 00:21:59.039
But what he means by that is well, um, you know, in general, the singularity, right?
00:21:59.359 --> 00:22:18.240
So the Singularity is kind of like a hypothetical future point, uh, when AI surpasses human intelligence, but also it begins to recursively improve itself, and then it triggers or could trigger an uncontrollable explosive growth in technological progress.
00:22:18.559 --> 00:22:25.839
So I I don't know, by all intent, right, by everything that's going on, you can make an argument, right?
00:22:25.920 --> 00:22:37.759
Like a lot of uh, you know, naysayers or people um you know that kind of rally against you know what Sam Altman or any of the big AI CEOs say, you know, a lot of people say, Oh, you know, they're just saying this for marketing.
00:22:37.920 --> 00:22:46.720
But I don't know when you look at what's been happening, like I've said, over the last eight weeks, I wouldn't tend to agree, right?
00:22:47.039 --> 00:22:48.799
Are we in this singularity?
00:22:48.960 --> 00:23:02.079
Maybe, or at least maybe this is the uh the beginning, and we're getting into that phase where yes, we are at the point now uh where AI systems are well way smarter than humans can even comprehend.
00:23:02.240 --> 00:23:05.200
Um, and they're beginning to improve themselves.
00:23:05.359 --> 00:23:06.960
And when that happens, right?
00:23:07.119 --> 00:23:20.160
Um, kind of once RSI is full RSI, no human, that's when the pace of acceleration um gets so fast that even the people building it, you know, can't even understand or keep up.
00:23:20.319 --> 00:23:26.559
And I do think that, you know, maybe we are starting to enter uh that era of being in the singularity, right?
00:23:26.720 --> 00:23:28.240
Are we definitively there?
00:23:28.480 --> 00:23:29.200
Maybe not.
00:23:29.440 --> 00:23:31.680
Are we entering into that frame?
00:23:31.839 --> 00:23:33.039
I would say so, right?
00:23:33.119 --> 00:23:39.519
Like I've been the firm believer we've been, you know, well past the you know artificial general intelligence um era.
00:23:39.759 --> 00:23:45.599
Um, and we are probably more headed in this, you know, kind of the in-between uh phase, right?
00:23:45.680 --> 00:23:59.039
Which is this singularity, which is, you know, RSI, uh, you know, recursive self-improvement, leading to the point where, well, we are gonna see super intelligence probably, I don't know, in the next uh decade uh or less, right?
00:23:59.119 --> 00:24:00.880
I'd say that's probably a safe bet.
00:24:01.279 --> 00:24:04.640
So um it's it's a big claim that he said, right, on the podcast.
00:24:04.880 --> 00:24:18.079
Uh, but what he kind of meant by it is, you know, seeing this steady compounding progress where AI helps build the next AI, not necessarily the you know, Terminator Skynet scenario, but you know, here's kind of what exists today.
00:24:18.319 --> 00:24:23.680
So we don't have the full human-free RSI, right?
00:24:23.759 --> 00:24:24.960
And I want to make that clear.
00:24:25.039 --> 00:24:32.960
Uh, but because you're gonna be seeing a lot more about recursive self-improvement in the news as Anthropoc and Open AI go public, um, right?
00:24:33.039 --> 00:24:38.079
I'm sure they're gonna be timing their model releases around certain uh dates in terms of them going public.
00:24:38.160 --> 00:24:42.240
So there's gonna be a lot of confusion about, you know, what does all of this mean, right?
00:24:42.400 --> 00:24:44.799
What are the actual capabilities of these systems?
00:24:44.880 --> 00:24:49.359
But full RSI is where AI builds its successor, right?
00:24:49.519 --> 00:24:52.480
The next model, um, but it doesn't need humans.
00:24:52.559 --> 00:24:55.839
So maybe humans aren't overseeing it, um, right?
00:24:56.000 --> 00:25:05.440
So that piece hasn't happened yet, but we do know um that the big labs are using their current best models to help build, right?
00:25:05.519 --> 00:25:11.759
But it's probably a little bit more uh human oversight uh right now, or a little bit more human-led than full AI led.
00:25:11.839 --> 00:25:15.119
But Anthropic did say that that situation isn't inevitable.
00:25:15.279 --> 00:25:16.240
So what's real today?
00:25:16.400 --> 00:25:20.880
Well, the AI does the improvement work while humans still pick the goals and improve the results.
00:25:21.200 --> 00:25:27.039
But the AI is being able to handle exponentially more powerful and longer task jobs.
00:25:27.200 --> 00:25:34.880
So um, according to meter, um, AI, the task length is doubling roughly every four months, right?
00:25:35.599 --> 00:25:49.759
Which, I mean, the projections um, you know, by like next year at this time, you know, that a single AI model is going to be able to do uh work that would normally take a human like multiple weeks, right?
00:25:49.920 --> 00:25:57.359
Which is crazy to think about at a uh, I forgot if that's the 50% or the 80% pass rate, but regardless, right?
00:25:57.519 --> 00:26:01.680
Whereas, you know, two years ago, we weren't even having this conversation.
00:26:01.839 --> 00:26:10.880
We weren't even thinking of of you know the point where a model could work for you know hours or days, or we've even seen sometimes more than a week, right?
00:26:10.960 --> 00:26:22.400
The standard out-of-the-box models can work on these hard problems agentically, pulling tools, uh you know, starting to, you know, uh investigate one path, rewind, and go down another.
00:26:22.559 --> 00:26:24.400
Um, so now think of what happens.
00:26:24.480 --> 00:26:30.559
Well, when the labs are using this, and right, we what you and I use, right?
00:26:30.640 --> 00:26:34.319
The the GPT-5-6 souls and the fable fives, right?
00:26:34.480 --> 00:26:36.319
These these are not the best models, right?
00:26:36.400 --> 00:26:38.240
Obviously, all the companies they have better models.
00:26:38.319 --> 00:26:48.720
So think of when the better models that they have are actually building and improving those own versions that we aren't having yet, right?
00:26:48.799 --> 00:27:12.960
And and my hope is that ultimately what this means is well, maybe this this token maxing uh era was just a whiplash of waiting for uh the good some of the good benefits, some of the good early benefits of recursive self-improvement, uh, right, which is maybe just cheaper and more affordable AI, uh, at least if you are with a company that has properly invested in compute.
00:27:13.359 --> 00:27:16.799
So let's go back to um the Google Deep Mind.
00:27:16.880 --> 00:27:28.000
I know we've touched on this a couple of times, but uh this was Google Deep Mind's uh Jazjeet Shakan, who said that RSI is now central to the AI industry's investment thesis, uh, right?
00:27:28.079 --> 00:27:32.960
So he said this at a conference, and he was comparing that build out to the Apollo program as well.
00:27:33.039 --> 00:27:42.559
Um, and even talking about, you know, with that, Alphabet is planning to spend up to$205 billion um this year in CapEx, right?
00:27:42.640 --> 00:27:52.079
So if you don't know CapEx, that's capital expenditures, that's essentially, you know, the the hardware uh side, the data centers, the you know, the chips, the cooling, all that.
00:27:52.160 --> 00:28:10.799
Um, but the twist is this it's the same uh companies or the employees from those companies that are investing um into you know these hundreds of billions of dollars into ultimately the data infrastructure that makes recursive self-improvement possible.
00:28:10.960 --> 00:28:17.279
It is the employees from those companies that are actually saying, wait, this whole RSI thing, we might want to pace this out.
00:28:17.440 --> 00:28:17.759
All right.
00:28:17.839 --> 00:28:23.759
So a little bit more on this letter that was called Pacing the Frontier that came out last week.
00:28:23.920 --> 00:28:27.039
So the signers are some of the biggest names in AI.
00:28:27.119 --> 00:28:36.240
I mean, they include Anthropic CEO, um, open AI's chief scientist, um, and more than 1,300 verified employees from the big labs.
00:28:36.319 --> 00:28:40.480
A lot of people from uh OpenAI and Anthropic make up the biggest pack.
00:28:40.640 --> 00:28:47.200
You get uh some people from Google, uh, Meta, DeepMind, Thinking Machines, everyone, right?
00:28:47.359 --> 00:28:50.799
And they aren't asking technically for a pause.
00:28:50.960 --> 00:28:58.000
Some of them kind of are in their individual comments, but the letter themselves, or sorry, the letter itself is not asking for a pause.
00:28:58.160 --> 00:29:11.359
It is asking for the US government to essentially um start to make a plan, uh, both um domestically and eventually internationally, to say, well, what happens and what are our plans?
00:29:11.519 --> 00:29:18.880
Um, you know, once the AI development is too fast for us, the people building it, to keep up with, right?
00:29:19.200 --> 00:29:30.000
And I do think this is probably um one of the, and you do have to tip your hat to uh, you know, these researchers and also the companies themselves, right?
00:29:30.079 --> 00:29:37.200
A lot of them put out statements saying, hey, we support, you know, we support this letter or we support, you know, our employees who who sign this letter.
00:29:37.359 --> 00:29:42.160
Um, you know, some people look at this as you know being anti-accelerating, right?
00:29:42.240 --> 00:29:47.119
You know, there's been this, you know, this huge uh AI acceleration movement over the past few years, right?
00:29:47.279 --> 00:29:49.359
Accelerate at all costs, uh, right.
00:29:49.440 --> 00:29:55.200
You know, this can help us, you know, it's you know, get to this utopian, potential utopian future, right?
00:29:55.279 --> 00:29:58.079
But then there's the potential dystopian future as well, right?
00:29:58.240 --> 00:30:10.880
Each coin has two sides to it, and you can't just be focused on the utopian side uh without acknowledging the fact, well, there's a very dystopian side of what could happen if RSI or AGI ASI goes incredibly wrong.
00:30:11.039 --> 00:30:11.359
All right.
00:30:11.440 --> 00:30:18.000
So these in this letter, they're not asking the government to stop AI, they just want the tools to be ready to potentially slow down later.
00:30:18.160 --> 00:30:20.960
Um, but is it possible?
00:30:21.119 --> 00:30:22.160
Probably not, right?
00:30:22.319 --> 00:30:25.920
Um, I talked about this um on yesterday's show.
00:30:26.079 --> 00:30:30.559
I don't think you know China is going to be like, yes, we're gonna sign this, right?
00:30:30.720 --> 00:30:41.839
Ultimately, um, you know, even though China is still probably two months behind uh with their more open source open weight models approach, they're they have the next models ready as well, right?
00:30:41.920 --> 00:30:54.160
We've seen impressive releases from Kimi K38, or sorry, uh Kimi, uh Kimi K3, uh Quen 3.8, uh GLM5.2, you know, reportedly GLM53 is on the way, right?
00:30:54.559 --> 00:31:04.720
I get I get the need, um, and and I very much so respect uh, you know, because it takes a lot of guts to put your sign your name on a paper like this, um, on a letter like this.
00:31:04.960 --> 00:31:12.480
But at the same time, I I give China a very low likelihood of um actually slowing down their pace.
00:31:12.640 --> 00:31:32.960
So it is good to have the foundations in place when and if things may go awry, and then the very rare case that the US and China on something as important as artificial intelligence, which I think is gonna become more important uh, you know, ultimately than any natural resources, gold, oil, uh, it's gonna become more important than you know, companies military.
00:31:33.039 --> 00:31:39.279
Uh, you know, controlling AI ultimately means you take the driver's seat for being the global superpower.
00:31:39.359 --> 00:31:42.240
So I don't think China's gonna slow down, anyways.
00:31:42.319 --> 00:31:48.000
Um, I think the biggest worry though from those signing the letters is well, who's gonna ultimately check AI's work?
00:31:48.160 --> 00:31:51.839
Uh, a couple quotes that I thought were relevant from that.
00:31:52.000 --> 00:31:54.240
Um, so I wanted to read three of these.
00:31:54.400 --> 00:31:55.359
I think I have three of them up.
00:31:55.519 --> 00:31:55.920
Yeah, I do.
00:31:56.000 --> 00:31:57.039
Oh no, just two.
00:31:57.359 --> 00:32:12.319
Okay, so one was from uh hopefully these names right, uh Elena Slocum, a member of technical staff at Anthropic, who said automated AI research is a technology that will profoundly alter the course of human history, for better or for worse.
00:32:12.480 --> 00:32:18.640
It is imperative that we coordinate our efforts to ensure this technology becomes a boon for all mankind.
00:32:18.799 --> 00:32:29.119
Uh, and then uh Sheng Jai Zhao, the chief scientist at Meta, said this AI is progressing at a rate that our society might not be ready for.
00:32:29.279 --> 00:32:36.480
Frontier labs are very close to an AI that can exceed even the best people on almost every metric of intelligence.
00:32:36.640 --> 00:32:40.720
This will lead to unprecedented social and safety risks.
00:32:40.880 --> 00:32:46.160
To ensure a positive future, we need to develop AI in a way that is driven by responsibility and thoughtfulness.
00:32:46.480 --> 00:32:51.200
So, um here's here's my my blunt takeaway from this.
00:32:52.000 --> 00:33:00.160
Obviously, these researchers, scientists know far more about the capabilities than we do today, right?
00:33:00.240 --> 00:33:06.640
We're we're essentially living in archaic times already, even with today's best publicly available models.
00:33:06.799 --> 00:33:20.000
Because not only, right, I've talked about it, we're probably gonna see uh uh GPT-5.7 or a GPT-6, you know, next month, we're probably gonna be seeing a Fable 5.1 fairly soon, you know, maybe even in August.
00:33:20.160 --> 00:33:20.880
We'll see.
00:33:21.039 --> 00:33:21.759
Uh right.
00:33:21.920 --> 00:33:36.079
And and these companies are probably already working on the next version that comes after this, because my assumption is um a lot of those models are you know post-training and there's probably already you know new runs starting on the next models, anyways.
00:33:36.319 --> 00:33:41.359
They know what's coming next because they're working on it, they're researching it, right?
00:33:41.519 --> 00:33:48.559
So I I think a lot of people, again, are saying, well, you know, hey, I'm just using Copilot to improve my emails.
00:33:48.799 --> 00:33:50.480
Like, what are these people talking about?
00:33:50.720 --> 00:33:51.759
They know, right?
00:33:51.920 --> 00:34:08.719
Even if you are in the bubble a little bit more and you've been able to uh kind of see how powerful today's systems are, uh, right, being able to automate a large part of your old job description, um, what's coming next seemingly is even more powerful and potentially more worrisome.
00:34:08.880 --> 00:34:09.840
So here's the catch.
00:34:09.920 --> 00:34:13.599
I think that self-improving AI needs a trustworthy scorekeeper.
00:34:13.840 --> 00:34:28.079
And I think that's the crux of what this um pacing the frontier uh letter was ultimately getting at, you know, that a self-improving AI that gets better, you know, at whatever it's scoring system rewards, even if it's wrong, right?
00:34:28.159 --> 00:34:31.840
And that's where we're kind of seeing uh these containment breaks, right?
00:34:31.920 --> 00:34:33.679
Because these AI models, right?
00:34:33.760 --> 00:34:39.920
If if you say, hey, you're gonna be rewarded for doing the best on this benchmark, and then it breaks the containment, right?
00:34:40.000 --> 00:34:43.119
And it's saying, well, I'm just doing my job, right?
00:34:43.199 --> 00:35:01.519
So, you know, you need a better way that, you know, especially as we enter a more full era uh of recursive self-improvement to um, you know, reward these models and to make sure that there's uh more trustworthy systems in place, you know, during the training and post-training processes.
00:35:01.760 --> 00:35:11.440
So we've seen multiple instances of this, like I've said, but I think it's so far been relatively harmless agent outbreak over the past 10 days, both from uh open AI and anthropic.
00:35:11.599 --> 00:35:16.239
But once RSI arrives, those kind of outbreaks could be catastrophic.
00:35:16.400 --> 00:35:20.000
Not, and I'm not saying from open AI and anthropic, that's not what I'm saying.
00:35:20.159 --> 00:35:21.920
I am saying ultimately, right?
00:35:22.000 --> 00:35:33.679
And I said this in yesterday's show, um, in in two years, right, on consumer hardware, you're gonna have models that are more capable than Fable 5.
00:35:33.760 --> 00:35:39.840
You're gonna have models that are more capable uh than GPT-5.6 soul on consumer desktops, right?
00:35:40.000 --> 00:35:40.960
Open source.
00:35:41.119 --> 00:35:43.920
Um, and that's where I start to worry about, right?
00:35:44.079 --> 00:35:59.760
When you have a version of an RSI model that you know you could be running on your computer, and that could be spitting out, you know, dozens or hundreds, who knows, of new smaller models a day that are super smart at super tasks, right?
00:35:59.840 --> 00:36:11.440
So that it's it's more of you know, when the guardrails could be let out, which is you know, open source does open that Pandora's box um of of you know how AI could be used in a bad way, right?
00:36:11.519 --> 00:36:21.039
At the same time, it does give defenders, I think, more access to more tools uh to be able to uh, you know, make the internet and hopefully the physical world a safer place.
00:36:21.199 --> 00:36:29.119
Um, but I think that humans are still gonna have to own that judgment role, even as RSI becomes more commonplace.
00:36:29.199 --> 00:36:33.519
And that's exactly where, as we wrap, I want your business strategy to start.
00:36:33.679 --> 00:36:36.480
So I threw a lot of information up uh at you.
00:36:36.559 --> 00:36:38.400
I wanted a couple tangents, right?
00:36:38.480 --> 00:36:41.679
That's why this thing's, you know, largely unedited, unscripted.
00:36:41.760 --> 00:36:47.039
I want you to get, you know, just kind of um some some real advice and and thoughts and feedback.
00:36:47.119 --> 00:36:49.840
But as we wrap up, what does all this mean for you?
00:36:50.159 --> 00:36:50.719
Right?
00:36:50.960 --> 00:37:07.360
Hey, if ultimately you're just trying to, you know, make your um, you know, agents more secure, uh, if you're just trying to increase productivity, if you're just trying to get more out of everyday, you know, you know, your um your AI systems that you have in place at your company.
00:37:07.519 --> 00:37:08.480
Why does this matter?
00:37:08.719 --> 00:37:09.679
Why does this matter?
00:37:09.840 --> 00:37:14.079
Well, you have to design for the loop, I think, or you're gonna get dragged.
00:37:14.239 --> 00:37:15.360
Here's what I mean by that.
00:37:15.519 --> 00:37:22.719
You have to plan, I think, for AI prices to actually fall, which you might have just gotten wind on this.
00:37:22.800 --> 00:37:25.760
Oh my gosh, we need to cut back on token spent.
00:37:25.920 --> 00:37:44.480
I think we have to look past that because the labs that have properly dedicated enough compute um to maybe achieve um RSI or, you know, uh what some labs are calling an automated AI researcher, those are the companies that, well, they're gonna be able to make prices cheaper.
00:37:44.559 --> 00:37:54.320
And we quite literally saw that um with OpenAI uh decreasing the price of GPT-5.6 Luna, a very capable model, right?
00:37:54.400 --> 00:37:57.199
You can't just say, oh, it's their smallest model, it's not any good.
00:37:57.280 --> 00:37:58.480
No, it's very good.
00:37:58.639 --> 00:38:05.119
Um, right, it is much better than you know, GPT-5, GPD 5.1, 2, 3, 4, right?
00:38:05.199 --> 00:38:09.920
It is a very capable model, and it is like free 99 right now.
00:38:10.079 --> 00:38:22.719
Um, so I think you have to prepare for prices to ultimately fall, but I think you ultimately have to build modularly because maybe you know you're with a different provider, and well, maybe your provider, right?
00:38:22.800 --> 00:38:30.159
At least when we look at the big four, the big six, maybe they're gonna figure out RSI a year after the whoever figures it out first.
00:38:30.320 --> 00:38:34.480
So that's why I think you can't be too entrenched uh with one provider.
00:38:34.639 --> 00:38:38.719
Um, I think you should be, you know, always choose your AI operating system of choice.
00:38:38.800 --> 00:38:39.679
I'm very transparent.
00:38:39.760 --> 00:38:45.039
I've always said for me and for most companies, I advise, I say that's Chad GPT because it's the easiest to learn, right?
00:38:45.119 --> 00:38:50.000
But whether you're anthropic, co-pilot, Gemini, Chad GPT, uh, Grok, Meta, right?
00:38:50.079 --> 00:38:56.639
I don't know anyone that any companies that are necessarily have chosen that as their AI operating system, but you should always be looking at fallbacks, right?
00:38:56.719 --> 00:39:01.119
Whether that's another frontier proprietary uh provider, whether that's an open model.
00:39:01.280 --> 00:39:04.079
Um, but you should also be copying the labs, right?
00:39:04.239 --> 00:39:08.639
They're using their expensive models for the hardest tasks only, right?
00:39:08.719 --> 00:39:11.760
And then they're using uh, you know, other models, right?
00:39:11.840 --> 00:39:16.719
So they're using that for judgment and planning, but then they're using cheaper models for the busy work, and you should be doing that too.
00:39:16.960 --> 00:39:17.760
I've already said this.
00:39:17.840 --> 00:39:31.280
We're at the point now with the capability overhangs where you don't need a Fable Five and a GPT 5.6 soul to rewrite your emails or to do the equivalent of a tough Google search, right?
00:39:31.360 --> 00:39:34.559
Or a simple uh you know, looking up some facts.
00:39:34.639 --> 00:39:39.920
Um, you know, you have to have um that kind of separated.
00:39:40.079 --> 00:39:42.880
Um, also another way to start copying the labs.
00:39:42.960 --> 00:39:45.039
This is something I've been doing now for a couple of months.
00:39:45.280 --> 00:39:52.880
Well, is you should have your AIs control other AIs and you should uh right, you maybe start building other AIs, right?
00:39:52.960 --> 00:39:57.840
And I think we didn't really get into that uh that realm until we got to GPT 5.6 soul, right?
00:39:57.920 --> 00:40:08.400
You had some people share online, and I talked about in the show about how people with no background in creating models were able to create, you know, small models using GBT5.
00:40:12.559 --> 00:40:19.360
Is someone used uh GPT 5.6 soul to look in, look at every single text message that they had ever sent via iMessage, right?
00:40:19.440 --> 00:40:31.360
So they had soul work on this and they created a literal small language model with weights um that was just based on here's you know everything I text about, here's how I write, here's and you know, and it knows everything about me, right?
00:40:31.440 --> 00:40:32.800
So, so think of that, right?
00:40:32.960 --> 00:40:38.880
Think of how when when prices I think are ultimately going to be driven down as the models get more powerful.
00:40:39.039 --> 00:40:41.519
What are those moves that you can make and copy from the big labs?
00:40:41.599 --> 00:40:46.880
And then last but not least, the question isn't whether about AI builds AI.
00:40:47.039 --> 00:40:51.679
The question isn't about, you know, oh my gosh, what does that mean, you know, recursive self-improvement?
00:40:51.840 --> 00:40:53.199
You know, what does that mean for our business?
00:40:53.360 --> 00:40:54.559
Oh my gosh, freak out.
00:40:54.719 --> 00:40:57.280
No, it's just whether your business is ready.
00:40:57.519 --> 00:41:02.559
So you have to start thinking and talking openly about the implications, right?
00:41:02.719 --> 00:41:15.440
What happens, you know, when there's maybe dozens of AI models that are um distinct and made for your job, for your department, for your sector, because that time is coming.
00:41:15.599 --> 00:41:30.320
Because I think ultimately what recursive self-improvement means, aside from hopefully uh, well, better models coming out faster, but also cheaper prices and probably more specific in niche uh models that serve specific verticals.
00:41:30.400 --> 00:41:38.320
So you have to start replanning now in whatever systems that you've got in place in 2024, 2025, etc., those can't be long-standing systems.
00:41:38.400 --> 00:41:42.880
I talked, you can't be planning a year at a time, you should be planning a month at a time.
00:41:42.960 --> 00:41:48.400
And I know that's you know, uh kind of daunting for traditional digital transformation, but I think that's where you have to be at.
00:41:48.559 --> 00:41:48.880
All right.
00:41:48.960 --> 00:42:04.480
I hope this show was helpful going over a lot of these terms, RSI, but hopefully now you know a little bit more about recursive self-improvement, what it means when these models are kind of improving themselves or working on the next version of themselves or making AI maybe cheaper and faster and better for all of us.
00:42:04.639 --> 00:42:11.599
So now you know what happens when RSI and models start improving themselves and hopefully what it means for your business.
00:42:11.840 --> 00:42:25.039
If this was helpful, please consider reposting this if you're listening on social media, if you're listening on the LinkedIn machine, uh, and if you're listening to the podcast on Spotify or Apple Music, please subscribe, then go to your everydayai.com.
00:42:25.199 --> 00:42:26.079
So thank you for tuning in.
00:42:26.159 --> 00:42:29.039
Hope to see you back tomorrow and every day for more everyday AI.
00:42:29.119 --> 00:42:30.159
Thanks, y'all.
