WEBVTT
00:00:02.640 --> 00:00:12.080
A few weeks ago, the United States government said the quiet part out loud when it comes to open source models, at least from China.
00:00:12.320 --> 00:00:26.320
That's because in April the White House sent out an official memo accusing China of using distillation to illegally copy American AI models to create cheaper domestic knockoffs.
00:00:26.559 --> 00:00:31.440
And that declaration is really nothing new if you've followed AI for years.
00:00:31.600 --> 00:00:44.479
However, the recent distillation trend has completely reshaped one important landscape of enterprise AI: the decision between using open source models versus proprietary closed models.
00:00:44.799 --> 00:00:55.119
And in 2026, at least, it can actually be a tough choice between saving potentially millions of dollars versus running up your legal liability.
00:00:55.359 --> 00:01:09.439
About two years ago, before Chinese distillation was commonplace, there was a sizable gap between frontier models and open source AI models, or those models that you can essentially download or use for close to free.
00:01:09.680 --> 00:01:20.560
But now the gap is all but closed, which has thrust the open source versus closed source question into every enterprise boardroom in 2026.
00:01:20.719 --> 00:01:33.280
And although there's no one size fits all answer, we're going to be tackling the toughest topics and the most important takeaways as we take a zoomed-out view of open source models on today's show.
00:01:33.439 --> 00:01:46.879
That's why we're going over open source AI 101, why local models, cheap APIs, and AI agents change everything about making AI decisions in 2026 as part of our Start Here series.
00:01:47.200 --> 00:01:47.840
All right.
00:01:48.079 --> 00:01:50.000
Welcome to Everyday AI.
00:01:50.159 --> 00:01:56.319
Before we dig in, let's first zoom out and talk about the big picture here when it comes to open source AI.
00:01:56.560 --> 00:02:02.239
That's because, well, it's actually a legitimate thing now, right?
00:02:02.560 --> 00:02:07.599
Two years ago, enterprise companies weren't saying, let's use an open source AI model in production.
00:02:07.760 --> 00:02:09.520
Today it's actually happened.
00:02:09.599 --> 00:02:10.719
Uh, it's happening.
00:02:10.800 --> 00:02:27.840
That's because you know, maybe the most powerful open source models are only about two to six months behind frontier models, but it on even consumer hardware, you can be running essentially frontier level AI models from like just over 12 months ago.
00:02:28.000 --> 00:02:37.919
And the Chinese labs now distilling US models have kind of crashed the open API prices to pennies, right?
00:02:38.080 --> 00:02:49.520
So uh, yeah, not everyone out there on you know consumer or prosumer hardware can run the most powerful open source models, although I do think Google has something to say about that.
00:02:49.680 --> 00:03:11.280
But the most powerful open source models run for a fraction of the actual cost if you can't afford to run them locally, which has completely shifted the paradigm when it comes to enterprises making decisions on well, are we going to use a model from one of the big three open AI, Google, or Anthropic, or are we going to use a Chinese open source model and pay for it that way?
00:03:11.439 --> 00:03:30.240
And well, what this has also led to in 2026 is essentially 24-7 local agents that can run uh and also without costing a ton, and now having to actively and almost aggressively, uh aggressively go through kind of an AI cost triage.
00:03:30.400 --> 00:03:37.840
But going full open source does strip away the legal protection that the closed models often include.
00:03:39.039 --> 00:03:44.960
So stick with me for 25-ish minutes on today's Start Here series show.
00:03:45.039 --> 00:03:46.479
And here's what you're gonna learn.
00:03:46.639 --> 00:03:51.039
You're gonna know why the open versus closed source AI default just officially flipped.
00:03:51.199 --> 00:03:58.560
You're gonna know how Gemma 4 from Google puts year-old frontier capability on your laptop.
00:03:58.719 --> 00:04:11.840
You're gonna understand the two payoffs already shaping and reshaping how individuals and enterprises run AI and the hidden legal trade-off most executives miss when going fully open source.
00:04:12.080 --> 00:04:12.879
Let's get into it.
00:04:12.960 --> 00:04:14.240
My name's Jordan Wilson.
00:04:14.319 --> 00:04:17.040
Welcome to Everyday AI's Start Here series.
00:04:17.199 --> 00:04:21.120
This is the essential podcast series to learn the AI basics.
00:04:21.279 --> 00:04:26.639
And if you're an AI expert, this is your chance to freshen up and double down on your AI knowledge.
00:04:26.879 --> 00:04:29.120
Why do we start this Start Here series?
00:04:29.279 --> 00:04:36.160
Well, after 750 plus podcasts, I never really had a good answer when someone was like, Where do I start?
00:04:36.319 --> 00:04:37.839
What podcast do I start with?
00:04:38.000 --> 00:04:40.000
That's why we created the Start Here series.
00:04:40.160 --> 00:04:42.399
It's best, I think, if you listen in order.
00:04:42.560 --> 00:04:45.839
I think this is now volume 24 of the Start Here series.
00:04:45.920 --> 00:04:49.680
So maybe we'll wrap it up at 25, maybe we'll wrap it up at 30.
00:04:49.839 --> 00:04:50.480
I'm not sure.
00:04:50.560 --> 00:04:53.439
But the whole point of this is you can go to starthere series.com.
00:04:53.600 --> 00:04:57.439
That's going to give you free access to our exclusive inner circle community.
00:04:57.600 --> 00:05:00.879
And in the Start Here Series space, we make it even easier for you.
00:05:01.040 --> 00:05:02.959
So you can actually go listen.
00:05:03.040 --> 00:05:11.360
Uh, we have a Spotify playlist ready for all of the different Start Here Series shows, as well as a breakdown on each individual episode, all in one place.
00:05:11.439 --> 00:05:17.600
So make sure you go to starthearseries.com for exclusive access to that inside of our inner circle community.
00:05:17.759 --> 00:05:18.000
All right.
00:05:18.079 --> 00:05:22.319
And if you missed our last Start Here Series show, that was volume 23.
00:05:22.399 --> 00:05:29.920
We talked about headless software and why companies are building software for AI agents and not humans and well, what that means.
00:05:30.160 --> 00:05:37.040
So today in volume 24 of the Start Here series, we're going over open source AI 101.
00:05:37.600 --> 00:05:43.199
So here's the reality: closed AI used to be the de facto, right?
00:05:43.360 --> 00:05:55.279
And I I mean, honestly, there was really never even much of a discussion about open source AI in the enterprise, maybe until uh 2025, at least, not serious enterprise companies.
00:05:55.439 --> 00:05:59.120
Uh now it's a it's a real conversation, right?
00:05:59.199 --> 00:06:04.560
So it's no longer uh, you know, hey, we're just gonna choose whichever API works best for us, right?
00:06:04.639 --> 00:06:07.839
Whether that's open AI, Anthropic, um, or Google.
00:06:08.079 --> 00:06:14.480
Now most companies are looking at some of the open source alternatives, most of them coming from China.
00:06:14.639 --> 00:06:25.759
And the big kind of thing here is companies are starting to standardize around one frontier vendor uh in 2025 before this happened, and then they called it an AI strategy.
00:06:26.000 --> 00:06:38.079
And the assumption originally was well, that worked for three years until two very specific forces broke the standard paradigm when it came to open source models.
00:06:39.120 --> 00:06:47.279
First, the proprietary versus closed gap capabilities just completely changed.
00:06:47.519 --> 00:06:51.199
All right, so we talk about arena on here a lot.
00:06:51.279 --> 00:06:55.120
Uh for uh previously had a different name, now it's just arena.
00:06:55.600 --> 00:07:02.480
So you put in a prompt, you don't know the outputs, uh, you know what models they're from, and you vote for which one is better, right?
00:07:02.639 --> 00:07:04.959
So all these different models get an ELO score.
00:07:05.120 --> 00:07:13.600
And to really zoom out for our non-technical audience, because I know a lot of you in the Start Here series are not technical, I'd probably even say what an open source model even is, right?
00:07:13.680 --> 00:07:28.959
So uh the very simplified version is certain companies can release models open source under like an MIT uh or Apache 2.0 license, and that gives people the ability to download these actual models and to run them locally on your machine.
00:07:29.120 --> 00:07:32.160
And that is, well, one of the big trade-offs, uh, right.
00:07:32.240 --> 00:07:37.360
So you're not sending any private or potentially proprietary data uh in the cloud at all.
00:07:37.439 --> 00:07:39.519
Everything runs locally on your machine.
00:07:39.600 --> 00:07:41.040
So number one, it's private.
00:07:41.279 --> 00:07:43.920
Number two, it's well, free, right?
00:07:44.000 --> 00:07:51.920
Uh and then there are open source models that if you can't download them on your you know computer, because not everyone can, some of them are much larger.
00:07:52.000 --> 00:07:58.879
Uh, you can still essentially run those in the cloud for a fraction of the cost of what it would uh cost to run a proprietary model.
00:07:58.959 --> 00:08:02.879
So essentially, open source models are ones that you can download, you can modify.
00:08:03.040 --> 00:08:05.519
In some instances, you can even build products on top of it.
00:08:05.680 --> 00:08:14.480
All right, anyways, uh right until late 2025, there was a monstrous gap in the arena scores, right?
00:08:14.560 --> 00:08:25.040
So these ELO scores, when you put in the same prompt, you look at two outputs, everyone overwhelmingly always chose the best front, uh, the best frontier closed source model.
00:08:25.120 --> 00:08:27.199
And that really started to change, right?
00:08:27.279 --> 00:08:32.240
So the gap between the ELO scores, uh, well, it cut down by about 90%.
00:08:32.720 --> 00:08:38.320
So it went from about a 250-point gap from the best uh frontier or closed source model.
00:08:38.399 --> 00:08:41.440
Um, well, now it's only about 30 points, right?
00:08:41.600 --> 00:08:43.840
Give or take, depending on the day, right?
00:08:44.000 --> 00:08:49.440
But it's even uh, you know, recently a couple months ago, it was like 15 points, right?
00:08:49.600 --> 00:08:55.840
So at that point, you really have to be an AI expert to be able to uh decipher the difference.
00:08:56.000 --> 00:09:09.039
I think 30 points, you know, most people could look at different outputs over time, and you know, a 30 point you can kind of realize it if you're looking at the best, you know, open source model versus the best proprietary closed models, 30 points you can understand.
00:09:09.200 --> 00:09:15.840
But 10 to 15 points, it's kind of a coin flip, even for people who are you know spending most of their days inside of large language models.
00:09:16.080 --> 00:09:23.600
Um, but this collapse came from those two different forces working in parallel.
00:09:24.159 --> 00:09:29.279
So I have a little uh little graphic here on the screen for our live stream audience.
00:09:29.360 --> 00:09:35.039
If you are listening on the podcast FYI, you can always get the video version on our website at your everydayai.com.
00:09:35.120 --> 00:09:53.279
Uh, but going from a 250-point gap to essentially a 30-point gap, uh this is huge, uh, right, because like I said, in 2023 to mid-2025, it was noticeable, right?
00:09:53.679 --> 00:10:02.240
It was extremely when you looked at the outputs, you could say my business can use output A, but it cannot use output B, right?
00:10:02.399 --> 00:10:11.200
And now we're at the point where the open source models, uh, in terms of an ELO score, and I think that's a good metric to look at over time, right?
00:10:11.360 --> 00:10:14.559
Because you know, the frontier is always improving.
00:10:14.639 --> 00:10:27.120
Uh, but if you look at the ELO scores of the open source models now, so those that you can kind of, you know, if you have a beefy enough computer, you can download some of the best open source models on your uh actual local machine, um, right?
00:10:27.440 --> 00:10:32.480
Those scores are where we were at with proprietary models three to six months ago, right?
00:10:32.639 --> 00:10:43.279
So think back to the very end of 2025, you know, and there's some models, like I think at the time was probably uh GPT-5.3, uh Gemini 3 Pro.
00:10:43.759 --> 00:10:48.960
And uh at that at that time, I think we were at like Opus uh maybe four or six or maybe four or five, right?
00:10:49.120 --> 00:10:56.559
Now you have open source models that you can run for free 24-7, run agentically, that are at that same level.
00:10:56.720 --> 00:11:07.679
And that's why now this is a real enterprise boardroom problem, especially for large companies that have invested heavily uh into AI, right?
00:11:07.759 --> 00:11:18.879
So I'm not talking about companies that, you know, with a couple hundred employees, I'm talking about companies that were spending seven, maybe eight, maybe even more, seven to eight figures on AI each year.
00:11:18.960 --> 00:11:28.799
Now all of a sudden they're saying, hey, in theory, if we switch, you know, part of our uh, you know, summarization tasks alone, right?
00:11:28.879 --> 00:11:33.200
Uh there's I've I've read a lot, uh, I've talked with a lot of people that have done something similar.
00:11:33.440 --> 00:11:42.080
You know, if if if we just you know chunk off everything that we're using, you know, open source model or sorry, closed source model, just for summarizing text, right?
00:11:42.159 --> 00:11:47.840
Some of those lower hanging fruit, people are saying, well, yeah, we could save one, two, three, four million dollars.
00:11:48.000 --> 00:11:52.720
And this is an actual reality that a lot of companies are uh grappling with right now.
00:11:52.799 --> 00:11:57.919
So the force one was just the capability moving locally, right?
00:11:58.480 --> 00:12:11.279
Uh and this, I think we have to credit Google for pushing the edge of edge AI.
00:12:11.519 --> 00:12:20.799
That's because with their Gemma 4 model, it completely shook up uh the landscape of open source AI, right?
00:12:20.960 --> 00:12:28.080
This thing was 20 times more efficient than other open source AI models at the time.
00:12:28.159 --> 00:12:30.480
So uh essentially, let me describe it like this.
00:12:30.639 --> 00:12:32.240
Do you remember GBT 4.0?
00:12:32.559 --> 00:12:36.879
Right, one of the best models, you know, about 14-15 months ago.
00:12:37.120 --> 00:12:40.000
It was at the absolute frontier, right?
00:12:40.159 --> 00:13:01.200
So now you can download Gemma 4 on a consumer laptop, and it has, you know, roughly you look at the scientific benchmarks and the ELO score, it's essentially about a GPT-4.0 level model that you can run on your laptop.
00:13:01.360 --> 00:13:02.320
So, what's the big deal?
00:13:02.399 --> 00:13:03.759
What's the big difference, right?
00:13:04.320 --> 00:13:17.600
14-15 months ago, I mean, there's thousands of companies spending millions of dollars a year to get that type of technology, to get GPT-4.0 level technology for their employees.
00:13:17.919 --> 00:13:20.639
Now it doesn't take take anything, really.
00:13:20.799 --> 00:13:23.840
It takes a new-ish piece of Apple, right?
00:13:23.919 --> 00:13:26.159
I I just got a new MacBook Pro.
00:13:26.320 --> 00:13:30.159
That thing can run Gemma 4 very easily, right?
00:13:30.399 --> 00:13:32.720
Can run even better models than that.
00:13:32.879 --> 00:13:41.120
And and this really changes, I think, what is ultimately capable when you look at the open source versus closed source.
00:13:41.360 --> 00:13:48.399
Because yes, I think most people look at normal usage and they're comparing apples to apples, right?
00:13:48.559 --> 00:13:54.159
Here's what our marketing team did 15 months ago with a GPT-4.0 level model.
00:13:54.320 --> 00:13:57.360
Oh, now they can do that on Gemma 4.
00:13:57.519 --> 00:14:11.840
Well, yeah, you can do it, but now you can do it agentically because not only in the last year or so have the models obviously improved with now thinking models, reasoning models being the default, but now we have these agentic harnesses.
00:14:12.000 --> 00:14:20.159
Um, you know, not just the ones that you can use um, you know, inside of Chat GPT, Gemini, uh, Claude Copilot.
00:14:20.320 --> 00:14:32.960
But well, you have these local um autonomous AI systems as well, such as OpenClaw, such as uh, I always forget if it's Hermes or Hermes agent, right?
00:14:33.120 --> 00:14:51.120
Um, so now you can have essentially the level of AI from 14 months ago running for free 24-7 agentically, even if you're just doing it, you know, summarization, content creation, um, research, things like that.
00:14:51.279 --> 00:15:05.519
So when capabilities went local, right, Gemma Ford leading the way, but obviously all the Chinese models followed suit because of uh right distillation, which we'll talk about a little bit here uh in a couple of minutes.
00:15:06.000 --> 00:15:13.840
But you can't overlook Gemma because it put frontier capability on literally a laptop, right?
00:15:14.159 --> 00:15:27.759
Because two years ago, to be able to run something like a GPT 4.0 level model, uh, right, which uh rumors have been swirling that's it's a two trillion parameter model, right?
00:15:28.080 --> 00:15:34.639
You would need a small little data center to run something like that, you know, two-ish years ago.
00:15:34.799 --> 00:15:36.960
Uh now you have these capabilities.
00:15:37.200 --> 00:15:41.919
So um, you know, I've been lucky to talk to a lot of smart people in AI.
00:15:42.000 --> 00:15:49.200
And now you really have executives grappling with, well, should we be buying a bunch of, as an example, new MacBooks?
00:15:49.440 --> 00:15:59.679
Should we be, you know, buying a bunch of DGX Sparks uh for our employees and setting them up with 24-7 always on agentic AI, right?
00:15:59.759 --> 00:16:06.320
To take advantage of these now local and powerful models that, well, you don't pay, right?
00:16:06.559 --> 00:16:11.919
You download them once, you don't pay again, and they work and they can work while you sleep.
00:16:12.080 --> 00:16:19.759
Like I said, because of uh some of the new autonomous capabilities uh from local agents that can run around the clock.
00:16:20.799 --> 00:16:30.799
So this has obviously led to, and I think uh Google putting the pressure on the uh the open source world with Gemma 4.
00:16:30.879 --> 00:16:37.360
Like I said, it was 20 times more efficient in terms of what is it was able to achieve on the benchmarks in terms of size, right?
00:16:37.519 --> 00:16:45.679
Because uh when it came out, if you looked at the other Chinese models, it was about 10 to 20 times smaller in size, right?
00:16:45.840 --> 00:16:59.200
So uh you wouldn't have been able to, you know, use the best open source model pre-Gemma 4 on a local machine, uh, right, at least a consumer laptop that you can just go walk into the store and buy.
00:16:59.360 --> 00:17:00.240
Now you can.
00:17:00.480 --> 00:17:11.039
And open Chinese models are amazing, and I think they've been getting better and better and smaller and smaller and more and more efficient since Google's Gemma 4.
00:17:11.200 --> 00:17:13.920
But you have to talk about the elephant in the room.
00:17:14.160 --> 00:17:18.000
That is, these models are distilled, right?
00:17:18.319 --> 00:17:27.119
We can say that all the big labs have said that, you know, are I don't know, maybe some of our uh audience in China won't appreciate hearing that.
00:17:27.279 --> 00:17:34.400
But I mean, Google, Anthropic, OpenAI have all accused China and have said they have proof, uh, right.
00:17:34.640 --> 00:17:36.880
But the uh the White House.
00:17:36.960 --> 00:17:52.480
So in April, the White House actually officially said that China was uh using uh kind of illegal uh tactics to distill uh US AI models to create cheaper domestic knockoffs, uh right.
00:17:52.559 --> 00:17:58.799
So it got to the point that at least the White House said that they had enough uh information or intel to make that declaration.
00:17:58.960 --> 00:18:02.720
So, what is model distillation and well, why does it matter?
00:18:03.279 --> 00:18:09.920
So the easiest way is like I can spend 10 hours studying for a test, right?
00:18:10.960 --> 00:18:12.319
Think back in the classroom.
00:18:12.400 --> 00:18:14.319
I can spend 10 hours studying for a test.
00:18:15.039 --> 00:18:20.559
Someone behind me can look over my shoulder and spend 10 minutes and get the exact same answers.
00:18:20.880 --> 00:18:23.440
That's kind of like what model distillation is, right?
00:18:23.599 --> 00:18:33.519
You have the big AI companies here in the US spending billions of dollars, right, on any single new, you know, model pre-training as an example.
00:18:33.680 --> 00:18:42.400
And essentially, uh, you have uh certain actors in China who will use the API.
00:18:42.480 --> 00:18:52.799
And you know, different companies have come out with different levels of um proof and say, okay, well, they're creating you know thousands of spoof accounts, more or less.
00:18:53.039 --> 00:18:59.359
They're putting in all these inputs and training it on our models' outputs versus training it themselves.
00:18:59.519 --> 00:19:01.519
So, yeah, just kind of copying the homework.
00:19:01.680 --> 00:19:14.480
So, what this has led to is China has been able to put out these open source models, uh, really technically just pushing the frontier of open source by allegedly just copying the best US models out there.
00:19:14.559 --> 00:19:20.079
And what this has led to is well, it's a crashing out at the bottom price of intelligence.
00:19:20.240 --> 00:19:30.640
So, Deep Seek V4, as an example, DeepSeek, one of those companies that many uh of the AI labs here in the US have um accused of model distillation.
00:19:30.880 --> 00:19:42.400
DeepSeq V4 Pro, one of their newer models, now lists their price at 43 cents uh per million token inputs and 87 cents per million token outputs.
00:19:42.559 --> 00:19:50.720
That's like more than 25 times cheaper than the premium uh you know uh closed source proprietary models.
00:19:50.799 --> 00:19:56.079
And that is the reality that a lot of boardrooms are looking at right now, right?
00:19:56.240 --> 00:20:16.480
To make that math easy, it's like, okay, if we're spending a thousand dollars per month per employee on the API side, uh, right, if you're sorry, if you're spending um, let's just say$40,000 a year, okay, we could be spending a thousand dollars a year if we switch over uh to an open source model as an example.
00:20:17.279 --> 00:20:24.000
So that's here's what that actual leads that what that actually leads to, right?
00:20:24.079 --> 00:20:32.079
Kind of the uh the model distillation leads to more powerful, cheaper open source models from China.
00:20:32.480 --> 00:20:39.200
And well, it leads to people using them, but not always knowing the ramifications, right?
00:20:39.279 --> 00:20:49.920
So obviously, uh Google doing things the right way, but I think with these Chinese models, they've become increasingly popular, even in the enterprise, uh, which is tricky.
00:20:50.079 --> 00:20:57.440
And I don't think that most executives are fully understanding some of the consequences of using open source models.
00:20:57.599 --> 00:21:09.359
But this has led to essentially having a workforce of always on assistance, and they've shifted from expensive special projects to well, that's just now the default operating model.
00:21:09.599 --> 00:21:19.680
So, you know, this has just allowed kind of these the this new swarm of agentic AI that couldn't have really have existed before.
00:21:19.839 --> 00:21:31.680
Number one, the technology and the harnessing wasn't there, but number two, you take out uh, you know, at least Gemma and the Chinese models uh that it have been accused of distillation.
00:21:31.839 --> 00:21:36.720
And your your your options, aside from those, aren't really that good.
00:21:36.880 --> 00:21:42.799
All right, we're gonna talk more here after we take a quick break for a word from our partners.
00:21:43.119 --> 00:21:51.039
Adobe just introduced an entirely new way to create, bringing the power and precision of its creative suite into one conversational experience.
00:21:51.279 --> 00:21:57.440
Meet Firefly AI Assistant, now live in the Adobe Firefly app, the all in one creative AI studio.
00:21:57.519 --> 00:21:59.680
Powered by Adobe's creative agent, Firefly.
00:22:00.079 --> 00:22:07.920
Apply AI Assistant lets you start with your vision, just describe what you want, and shape the outcome as it takes form with the assistant.
00:22:08.079 --> 00:22:21.759
The Assistant orchestrates multi-step workflows, drawing on 60 plus pro-grade tools across Adobe Creative Cloud apps, including Photoshop, Illustrator Premiere, Lightroom, Express, and more to help bring your ideas to life.
00:22:22.000 --> 00:22:34.559
You can also get started with Creative Skills, a growing library of pre-built workflows for common creative tasks like batch editing photos, creating mood boards, portrait retouching, and creating social variations.
00:22:34.720 --> 00:22:40.480
Every step the assistant takes is visible, so you can refine, redirect, or take over at any time.
00:22:40.720 --> 00:22:43.279
You stay in the driver's seat as the creative director.
00:22:43.519 --> 00:22:47.119
Adobe Firefly AI Assistant, now in public beta.
00:22:47.279 --> 00:22:50.319
See it today at firefly.adobe.com.
00:22:52.480 --> 00:23:08.640
Aside from always on AI agents, what this open source movement has led to is well, now enterprises may be moving away from having the one model fits all uh solution.
00:23:08.960 --> 00:23:22.000
So now, as an example, you might be able to put out a 100-agent swarm goes from you know$1,200 plus dollars on Opus to well, maybe like$60 some dollars on Deep Seek.
00:23:22.480 --> 00:23:34.880
And now essentially you um can look at AI as more of a triage or categorical categorization of which models to use for which tasks, right?
00:23:35.039 --> 00:23:38.400
Especially when you're talking about high volume operations.
00:23:38.799 --> 00:23:48.880
So things like when you're going through it in bulk, things like summarization, uh extraction, uh parsing PDFs, classification, right?
00:23:49.279 --> 00:24:00.000
Now so many even large enterprise companies are no longer doing that on the back end using the frontier US uh companies.
00:24:00.160 --> 00:24:10.799
Well, I mean, many still are, but you've already seen a big segment uh of those companies move to these open source or uh Chinese open source models.
00:24:11.359 --> 00:24:17.119
But if you're thinking right now, if you're like, wow, our bill's pretty high, our API bill, right?
00:24:17.200 --> 00:24:24.480
I'm not talking about on the front end, you know, the number of seats you have in Chat GPT Enterprise or you know in Gemini Enterprise or anything like that, right?
00:24:24.559 --> 00:24:30.079
I'm talking about back end, uh, all of these special projects that you have running via the API.
00:24:30.240 --> 00:24:46.160
So if if you're looking at your API bill and you're like, yeah, we're going through a lot, uh, or you know, hey, we're using you know Opus 4.7 and GPT-5.5 to run our agents, maybe we should be looking at you know Kimmy or Deep Seek or uh whatever it is.
00:24:46.319 --> 00:24:52.000
Before you do that, you have to know that there is a big trade-off.
00:24:52.640 --> 00:25:02.720
Because just because a model is free or open source uh or cheap-ish to run via the API, right?
00:25:02.799 --> 00:25:10.480
If you are uh you know running some of these open models uh via the API, there's still uh an expensive price to pay.
00:25:10.720 --> 00:25:25.519
And that price might be unknown at this point, but it can cost your company a lot more than maybe just using that closed proprietary AI would have costed you could have cost you via the API.
00:25:26.559 --> 00:25:36.240
That's because using open source strips away all of that legal protection that you probably overlook or take for granted.
00:25:36.720 --> 00:25:37.680
What do I mean?
00:25:37.839 --> 00:25:44.319
Well, when you're using anthropic, open AI, Microsoft, um, who did I forget?
00:25:44.480 --> 00:25:46.559
Microsoft, open AI, Google, Anthropic, right?
00:25:46.640 --> 00:25:53.200
When you're using those uh enterprise offerings, you have a level of legal protection, right?
00:25:53.279 --> 00:25:58.559
So as an example, if you use, right, I'm not gonna go through all the uh fine prints, right?
00:25:58.640 --> 00:26:05.759
But the four companies at the enterprise level all offer, you know, some sort of essential uh uh I won't call it insurance, right?
00:26:05.920 --> 00:26:07.519
But think of it kind of like that, right?
00:26:07.599 --> 00:26:18.240
Like, hey, if you use something produced by our systems and if you use it ethically and responsibly and with guardrails and it produces something that's not you know correct, there is some level of protection there, right?
00:26:18.480 --> 00:26:22.160
Which you don't get that with open source models, right?
00:26:22.240 --> 00:26:28.880
So, as an example, you know, Deep Seek, uh, you know, they use uh different MIT licenses, Apache 2.0.
00:26:29.119 --> 00:26:35.680
Uh yeah, essentially there's no warranty or non-infringement agreements, all right.
00:26:35.920 --> 00:26:43.519
So for regulated work and customer-facing output, you have to look at the trade-off.
00:26:43.680 --> 00:26:48.079
Yeah, you might save six, seven, who knows?
00:26:48.400 --> 00:27:01.839
Maybe eight uh figures by switching the bulk of some of your maybe agentic or bulk workloads, uh, you write, especially if you're a fortune, fortune 500, fortune 100 company.
00:27:02.000 --> 00:27:10.240
There's minimum seven, eight figures that you could, in theory, save by switching some of those heavier agency uh or you know, parsing.
00:27:10.319 --> 00:27:22.720
You know, I know parsing is a big one, shifting some of those workflows to open models, but you lose that legal protection that maybe you've had to rely on it before, maybe you haven't.
00:27:22.960 --> 00:27:26.480
But that one time that you would actually need it.
00:27:26.720 --> 00:27:34.880
And if you do switch over to open source, you have to understand those ramifications because at that point you're gonna actually be paying for it.
00:27:35.759 --> 00:27:51.440
So that gets us to the real question here as we get close to wrapping up because I don't want my takeaway here to be don't use open models, they're not safe, because that's not the takeaway.
00:27:52.000 --> 00:28:00.000
I think you need to start looking at your AI workflow like a triage, right?
00:28:00.240 --> 00:28:02.799
At least when it comes to back end tasks, right?
00:28:02.960 --> 00:28:06.720
Uh, front end, I've always been a firm believer, and I still am today.
00:28:06.880 --> 00:28:14.400
You need to pick your AI operating system of choice, whether that's co-pilot, chat GPT, Claude, Gemini on the front end.
00:28:14.480 --> 00:28:22.480
And that's where you should move, especially your non-technical people, should move the majority of their day-to-day knowledge work tasks, should be happening on the front end there.
00:28:22.640 --> 00:28:26.240
But you still have a multitude of back-end tasks.
00:28:26.559 --> 00:28:33.039
And I think you have to look at it like triaging in an emergency room, right?
00:28:34.160 --> 00:28:44.799
You wouldn't send your top neurosurgeon in when someone's having an allergic reaction to honey, right?
00:28:45.119 --> 00:28:46.079
You wouldn't do that.
00:28:46.799 --> 00:28:54.319
You would save that neurosurgeon for well, someone that needs a neurosurgeon.
00:28:55.440 --> 00:29:00.799
And I think that there's so many companies that haven't gone through the basics of this.
00:29:01.039 --> 00:29:18.240
For the most part, on the API side, they pick, well, one model and they say, All right, well, we have our AI operating system of choice, and then for everything else, as an example, we go to Sonnet 4.6 or we go to you know Gemini 3.1 Flash or whatever that model may be.
00:29:18.400 --> 00:29:20.319
And maybe that's the right model.
00:29:20.559 --> 00:29:30.000
Maybe those companies have done their due diligence and have vetted out their different use cases and have priced it out, and maybe that's the right move, but maybe it's not.
00:29:30.160 --> 00:29:37.039
Because I know from experience and talking to a lot of people, a lot of companies just choose whatever is on the cutting edge.
00:29:37.200 --> 00:29:47.039
And they say, Well, this is the best, so we're gonna pay for it because there is a push internally to use more AI, to use the best AI.
00:29:47.200 --> 00:29:50.880
We see all these new benchmarks, we want to make sure that we're taking advantage of it.
00:29:51.039 --> 00:29:59.200
Well, is that neurosurgeon gonna be able to, you know, properly diagnose the allergic reaction to someone eating honey?
00:29:59.279 --> 00:30:03.039
Well, yeah, probably, but it's gonna cost you a lot more.
00:30:03.279 --> 00:30:08.319
So you need to think about sending those high-volume, low-stakes work, right?
00:30:08.400 --> 00:30:18.880
Like summarization, research, content creation, maybe, uh, to cheaper open source models that you can either run locally uh or you know, running via the API for just cost efficiency.
00:30:19.119 --> 00:30:24.640
If there's essentially like no legal ramifications if you get something wrong, right.
00:30:24.720 --> 00:30:30.079
So if you're in a highly regulated sector, this is probably not the advice for you, uh, right.
00:30:30.240 --> 00:30:37.359
You shouldn't probably be using uh, you know, or just taking my advice on a whole lot of anything as truth.
00:30:37.440 --> 00:30:40.799
You always need to be vetting these things out for yourself, uh, right.
00:30:40.880 --> 00:30:57.680
But if if if if it is something relatively in a sector that's not highly regulated, where there's not a quote unquote a lot on the line, that's one of those instances where you need to say, can we shift some of our more expensive API workloads to an open source model?
00:30:57.839 --> 00:31:04.160
Uh, or if you need to run sensitive private workflows on self-hosted open models, that's another thing.
00:31:04.319 --> 00:31:14.000
I think that there's still, even to this day, even though I think there's plenty of reasons, you know, one thing I always ask companies when they're like, Oh, we don't, well, we don't run this through AI, right?
00:31:14.079 --> 00:31:15.200
Because it's sensitive data.
00:31:15.279 --> 00:31:17.680
And I'm like, okay, well, do you have a cloud provider?
00:31:17.759 --> 00:31:18.960
And they're like, Of course.
00:31:19.200 --> 00:31:21.839
It's like, okay, well, it's the same thing, more or less, right?
00:31:22.000 --> 00:31:25.279
As long as you take proper precautions, turn off model training and all that.
00:31:25.359 --> 00:31:31.519
It's it's more or less using the same level grade of security that you know, cloud uh uses.
00:31:31.759 --> 00:31:36.079
Anyways, there are still some things that companies won't even put on the cloud, right?
00:31:36.240 --> 00:31:52.640
Which I understand, but having these now uh extremely powerful open source models and extremely efficient open source models, now you can start running those private workloads, uh uh workloads or workflows on-prem, right?
00:31:52.799 --> 00:31:56.240
Or you know, self-hosted that you can fully control.
00:31:56.880 --> 00:32:21.759
And then you can reserve those more premium, uh, those more um uh high value, highly sensitive tasks, um, you know, that for those models that can reason and think and offer kind of that that that level of uh security and legal support that you don't get if you opt for open models instead.
00:32:22.640 --> 00:32:25.519
So, you know, have a nice little chart here.
00:32:25.599 --> 00:32:43.519
So maybe when you look at the cheap open APIs, you look at simple tasks like summarization, extraction, or classification uh for local, self-hosted open models, private workflows, great for that, running agents locally, having more control, uh, right.
00:32:43.680 --> 00:32:52.720
And then for the premium closed models, which are the ones that a lot of people are using on the front end, on the back end, you should still be using these for a lot of reasons, right?
00:32:52.880 --> 00:33:00.640
Uh, those that, well, carry a lot of business value, uh hard tasks that require reasoning.
00:33:00.799 --> 00:33:02.400
Uh, right, your final review.
00:33:02.559 --> 00:33:18.559
Maybe you do uh, you know, draft version either with a cheap API or local self-hosted open model, but or anything, you know, uh that requires customer-facing output should probably be going on that premium closed models for that level of protection.
00:33:18.640 --> 00:33:33.279
Like I said, you cannot overlook the hidden trade-off that these open licenses may disclaim warranty and non-infringements, where the enterprise offerings do usually include that I that IP uh indemnification.
00:33:33.440 --> 00:33:41.839
So for regulated work, that protection in almost all cases justifies the premium that you pay.
00:33:42.240 --> 00:33:49.519
However, as we wrap up here, let me just quickly encapsulate all of this.
00:33:50.559 --> 00:33:52.400
Local models aren't going anywhere.
00:33:52.799 --> 00:33:53.119
All right.
00:33:53.279 --> 00:34:07.839
And I actually think, especially as we uh officially welcome in the era of models that can improve themselves and create can create own versions, uh, you know, smaller versions of themselves, right?
00:34:07.920 --> 00:34:17.360
All the big companies have essentially said, you know, have hinted uh at RSI or you know, the fact that our big models make smaller versions of themselves.
00:34:17.599 --> 00:34:31.119
I think that we're gonna not only see a continued trend toward uh local open source models, I think we're gonna start seeing a lot of smaller models uh for very specific use cases.
00:34:31.280 --> 00:34:33.920
It's something I've been you know predicting now for multiple years.
00:34:34.079 --> 00:34:35.360
We started to see it slowly.
00:34:35.440 --> 00:34:41.599
I think it is going to pick up steam now that we're starting to get some hints of recursive self-improvement with these models.
00:34:41.840 --> 00:34:47.119
So your company has to be paying attention because this is a trend that is not going away.
00:34:47.360 --> 00:35:00.159
The open models are going to become more and more capable, they're gonna become faster, they're gonna become more efficient, and the options are going to start to become even greater, uh, right?
00:35:00.320 --> 00:35:13.760
Not just great general purpose open models that can run on consumer hardware like Gemma 4, uh, but small open models for very specific tasks that can be highly valuable for your company.
00:35:14.000 --> 00:35:26.480
So you have to understand uh the pros and the cons of these local models uh when you might use a cheap API and how this changes the agentic outlook for your company.
00:35:26.639 --> 00:35:30.639
So don't write them off just because you always want to use the latest and the greatest.
00:35:30.719 --> 00:35:39.920
Yes, you should do that, but don't send the neurosarch in to you know triage a basic thing happening in the waiting room.
00:35:40.159 --> 00:35:43.199
Send the right model at the right time for the right purpose.
00:35:43.360 --> 00:35:50.639
So I hope this was helpful as we recapped open source AI 101 as part of our Start Here Series.
00:35:50.719 --> 00:35:53.760
If this was helpful, number one, make sure you subscribe to the podcast.
00:35:53.840 --> 00:35:54.559
I'd appreciate that.
00:35:54.639 --> 00:35:57.280
But then make sure you go to starthere series.com.
00:35:57.440 --> 00:36:01.920
That's going to give you free access to our exclusive inner circle community.
00:36:02.079 --> 00:36:06.559
Right now, there's no other way to join except by going to startheseries.com.
00:36:06.639 --> 00:36:07.519
So make sure you do that.
00:36:07.679 --> 00:36:08.400
Thank you for tuning in.
00:36:08.480 --> 00:36:11.519
I hope to see you back tomorrow and every day for more everyday AI.
00:36:11.599 --> 00:36:12.639
Thanks, y'all.
00:00:02.640 --> 00:00:12.080
A few weeks ago, the United States government said the quiet part out loud when it comes to open source models, at least from China.
00:00:12.320 --> 00:00:26.320
That's because in April the White House sent out an official memo accusing China of using distillation to illegally copy American AI models to create cheaper domestic knockoffs.
00:00:26.559 --> 00:00:31.440
And that declaration is really nothing new if you've followed AI for years.
00:00:31.600 --> 00:00:44.479
However, the recent distillation trend has completely reshaped one important landscape of enterprise AI: the decision between using open source models versus proprietary closed models.
00:00:44.799 --> 00:00:55.119
And in 2026, at least, it can actually be a tough choice between saving potentially millions of dollars versus running up your legal liability.
00:00:55.359 --> 00:01:09.439
About two years ago, before Chinese distillation was commonplace, there was a sizable gap between frontier models and open source AI models, or those models that you can essentially download or use for close to free.
00:01:09.680 --> 00:01:20.560
But now the gap is all but closed, which has thrust the open source versus closed source question into every enterprise boardroom in 2026.
00:01:20.719 --> 00:01:33.280
And although there's no one size fits all answer, we're going to be tackling the toughest topics and the most important takeaways as we take a zoomed-out view of open source models on today's show.
00:01:33.439 --> 00:01:46.879
That's why we're going over open source AI 101, why local models, cheap APIs, and AI agents change everything about making AI decisions in 2026 as part of our Start Here series.
00:01:47.200 --> 00:01:47.840
All right.
00:01:48.079 --> 00:01:50.000
Welcome to Everyday AI.
00:01:50.159 --> 00:01:56.319
Before we dig in, let's first zoom out and talk about the big picture here when it comes to open source AI.
00:01:56.560 --> 00:02:02.239
That's because, well, it's actually a legitimate thing now, right?
00:02:02.560 --> 00:02:07.599
Two years ago, enterprise companies weren't saying, let's use an open source AI model in production.
00:02:07.760 --> 00:02:09.520
Today it's actually happened.
00:02:09.599 --> 00:02:10.719
Uh, it's happening.
00:02:10.800 --> 00:02:27.840
That's because you know, maybe the most powerful open source models are only about two to six months behind frontier models, but it on even consumer hardware, you can be running essentially frontier level AI models from like just over 12 months ago.
00:02:28.000 --> 00:02:37.919
And the Chinese labs now distilling US models have kind of crashed the open API prices to pennies, right?
00:02:38.080 --> 00:02:49.520
So uh, yeah, not everyone out there on you know consumer or prosumer hardware can run the most powerful open source models, although I do think Google has something to say about that.
00:02:49.680 --> 00:03:11.280
But the most powerful open source models run for a fraction of the actual cost if you can't afford to run them locally, which has completely shifted the paradigm when it comes to enterprises making decisions on well, are we going to use a model from one of the big three open AI, Google, or Anthropic, or are we going to use a Chinese open source model and pay for it that way?
00:03:11.439 --> 00:03:30.240
And well, what this has also led to in 2026 is essentially 24-7 local agents that can run uh and also without costing a ton, and now having to actively and almost aggressively, uh aggressively go through kind of an AI cost triage.
00:03:30.400 --> 00:03:37.840
But going full open source does strip away the legal protection that the closed models often include.
00:03:39.039 --> 00:03:44.960
So stick with me for 25-ish minutes on today's Start Here series show.
00:03:45.039 --> 00:03:46.479
And here's what you're gonna learn.
00:03:46.639 --> 00:03:51.039
You're gonna know why the open versus closed source AI default just officially flipped.
00:03:51.199 --> 00:03:58.560
You're gonna know how Gemma 4 from Google puts year-old frontier capability on your laptop.
00:03:58.719 --> 00:04:11.840
You're gonna understand the two payoffs already shaping and reshaping how individuals and enterprises run AI and the hidden legal trade-off most executives miss when going fully open source.
00:04:12.080 --> 00:04:12.879
Let's get into it.
00:04:12.960 --> 00:04:14.240
My name's Jordan Wilson.
00:04:14.319 --> 00:04:17.040
Welcome to Everyday AI's Start Here series.
00:04:17.199 --> 00:04:21.120
This is the essential podcast series to learn the AI basics.
00:04:21.279 --> 00:04:26.639
And if you're an AI expert, this is your chance to freshen up and double down on your AI knowledge.
00:04:26.879 --> 00:04:29.120
Why do we start this Start Here series?
00:04:29.279 --> 00:04:36.160
Well, after 750 plus podcasts, I never really had a good answer when someone was like, Where do I start?
00:04:36.319 --> 00:04:37.839
What podcast do I start with?
00:04:38.000 --> 00:04:40.000
That's why we created the Start Here series.
00:04:40.160 --> 00:04:42.399
It's best, I think, if you listen in order.
00:04:42.560 --> 00:04:45.839
I think this is now volume 24 of the Start Here series.
00:04:45.920 --> 00:04:49.680
So maybe we'll wrap it up at 25, maybe we'll wrap it up at 30.
00:04:49.839 --> 00:04:50.480
I'm not sure.
00:04:50.560 --> 00:04:53.439
But the whole point of this is you can go to starthere series.com.
00:04:53.600 --> 00:04:57.439
That's going to give you free access to our exclusive inner circle community.
00:04:57.600 --> 00:05:00.879
And in the Start Here Series space, we make it even easier for you.
00:05:01.040 --> 00:05:02.959
So you can actually go listen.
00:05:03.040 --> 00:05:11.360
Uh, we have a Spotify playlist ready for all of the different Start Here Series shows, as well as a breakdown on each individual episode, all in one place.
00:05:11.439 --> 00:05:17.600
So make sure you go to starthearseries.com for exclusive access to that inside of our inner circle community.
00:05:17.759 --> 00:05:18.000
All right.
00:05:18.079 --> 00:05:22.319
And if you missed our last Start Here Series show, that was volume 23.
00:05:22.399 --> 00:05:29.920
We talked about headless software and why companies are building software for AI agents and not humans and well, what that means.
00:05:30.160 --> 00:05:37.040
So today in volume 24 of the Start Here series, we're going over open source AI 101.
00:05:37.600 --> 00:05:43.199
So here's the reality: closed AI used to be the de facto, right?
00:05:43.360 --> 00:05:55.279
And I I mean, honestly, there was really never even much of a discussion about open source AI in the enterprise, maybe until uh 2025, at least, not serious enterprise companies.
00:05:55.439 --> 00:05:59.120
Uh now it's a it's a real conversation, right?
00:05:59.199 --> 00:06:04.560
So it's no longer uh, you know, hey, we're just gonna choose whichever API works best for us, right?
00:06:04.639 --> 00:06:07.839
Whether that's open AI, Anthropic, um, or Google.
00:06:08.079 --> 00:06:14.480
Now most companies are looking at some of the open source alternatives, most of them coming from China.
00:06:14.639 --> 00:06:25.759
And the big kind of thing here is companies are starting to standardize around one frontier vendor uh in 2025 before this happened, and then they called it an AI strategy.
00:06:26.000 --> 00:06:38.079
And the assumption originally was well, that worked for three years until two very specific forces broke the standard paradigm when it came to open source models.
00:06:39.120 --> 00:06:47.279
First, the proprietary versus closed gap capabilities just completely changed.
00:06:47.519 --> 00:06:51.199
All right, so we talk about arena on here a lot.
00:06:51.279 --> 00:06:55.120
Uh for uh previously had a different name, now it's just arena.
00:06:55.600 --> 00:07:02.480
So you put in a prompt, you don't know the outputs, uh, you know what models they're from, and you vote for which one is better, right?
00:07:02.639 --> 00:07:04.959
So all these different models get an ELO score.
00:07:05.120 --> 00:07:13.600
And to really zoom out for our non-technical audience, because I know a lot of you in the Start Here series are not technical, I'd probably even say what an open source model even is, right?
00:07:13.680 --> 00:07:28.959
So uh the very simplified version is certain companies can release models open source under like an MIT uh or Apache 2.0 license, and that gives people the ability to download these actual models and to run them locally on your machine.
00:07:29.120 --> 00:07:32.160
And that is, well, one of the big trade-offs, uh, right.
00:07:32.240 --> 00:07:37.360
So you're not sending any private or potentially proprietary data uh in the cloud at all.
00:07:37.439 --> 00:07:39.519
Everything runs locally on your machine.
00:07:39.600 --> 00:07:41.040
So number one, it's private.
00:07:41.279 --> 00:07:43.920
Number two, it's well, free, right?
00:07:44.000 --> 00:07:51.920
Uh and then there are open source models that if you can't download them on your you know computer, because not everyone can, some of them are much larger.
00:07:52.000 --> 00:07:58.879
Uh, you can still essentially run those in the cloud for a fraction of the cost of what it would uh cost to run a proprietary model.
00:07:58.959 --> 00:08:02.879
So essentially, open source models are ones that you can download, you can modify.
00:08:03.040 --> 00:08:05.519
In some instances, you can even build products on top of it.
00:08:05.680 --> 00:08:14.480
All right, anyways, uh right until late 2025, there was a monstrous gap in the arena scores, right?
00:08:14.560 --> 00:08:25.040
So these ELO scores, when you put in the same prompt, you look at two outputs, everyone overwhelmingly always chose the best front, uh, the best frontier closed source model.
00:08:25.120 --> 00:08:27.199
And that really started to change, right?
00:08:27.279 --> 00:08:32.240
So the gap between the ELO scores, uh, well, it cut down by about 90%.
00:08:32.720 --> 00:08:38.320
So it went from about a 250-point gap from the best uh frontier or closed source model.
00:08:38.399 --> 00:08:41.440
Um, well, now it's only about 30 points, right?
00:08:41.600 --> 00:08:43.840
Give or take, depending on the day, right?
00:08:44.000 --> 00:08:49.440
But it's even uh, you know, recently a couple months ago, it was like 15 points, right?
00:08:49.600 --> 00:08:55.840
So at that point, you really have to be an AI expert to be able to uh decipher the difference.
00:08:56.000 --> 00:09:09.039
I think 30 points, you know, most people could look at different outputs over time, and you know, a 30 point you can kind of realize it if you're looking at the best, you know, open source model versus the best proprietary closed models, 30 points you can understand.
00:09:09.200 --> 00:09:15.840
But 10 to 15 points, it's kind of a coin flip, even for people who are you know spending most of their days inside of large language models.
00:09:16.080 --> 00:09:23.600
Um, but this collapse came from those two different forces working in parallel.
00:09:24.159 --> 00:09:29.279
So I have a little uh little graphic here on the screen for our live stream audience.
00:09:29.360 --> 00:09:35.039
If you are listening on the podcast FYI, you can always get the video version on our website at your everydayai.com.
00:09:35.120 --> 00:09:53.279
Uh, but going from a 250-point gap to essentially a 30-point gap, uh this is huge, uh, right, because like I said, in 2023 to mid-2025, it was noticeable, right?
00:09:53.679 --> 00:10:02.240
It was extremely when you looked at the outputs, you could say my business can use output A, but it cannot use output B, right?
00:10:02.399 --> 00:10:11.200
And now we're at the point where the open source models, uh, in terms of an ELO score, and I think that's a good metric to look at over time, right?
00:10:11.360 --> 00:10:14.559
Because you know, the frontier is always improving.
00:10:14.639 --> 00:10:27.120
Uh, but if you look at the ELO scores of the open source models now, so those that you can kind of, you know, if you have a beefy enough computer, you can download some of the best open source models on your uh actual local machine, um, right?
00:10:27.440 --> 00:10:32.480
Those scores are where we were at with proprietary models three to six months ago, right?
00:10:32.639 --> 00:10:43.279
So think back to the very end of 2025, you know, and there's some models, like I think at the time was probably uh GPT-5.3, uh Gemini 3 Pro.
00:10:43.759 --> 00:10:48.960
And uh at that at that time, I think we were at like Opus uh maybe four or six or maybe four or five, right?
00:10:49.120 --> 00:10:56.559
Now you have open source models that you can run for free 24-7, run agentically, that are at that same level.
00:10:56.720 --> 00:11:07.679
And that's why now this is a real enterprise boardroom problem, especially for large companies that have invested heavily uh into AI, right?
00:11:07.759 --> 00:11:18.879
So I'm not talking about companies that, you know, with a couple hundred employees, I'm talking about companies that were spending seven, maybe eight, maybe even more, seven to eight figures on AI each year.
00:11:18.960 --> 00:11:28.799
Now all of a sudden they're saying, hey, in theory, if we switch, you know, part of our uh, you know, summarization tasks alone, right?
00:11:28.879 --> 00:11:33.200
Uh there's I've I've read a lot, uh, I've talked with a lot of people that have done something similar.
00:11:33.440 --> 00:11:42.080
You know, if if if we just you know chunk off everything that we're using, you know, open source model or sorry, closed source model, just for summarizing text, right?
00:11:42.159 --> 00:11:47.840
Some of those lower hanging fruit, people are saying, well, yeah, we could save one, two, three, four million dollars.
00:11:48.000 --> 00:11:52.720
And this is an actual reality that a lot of companies are uh grappling with right now.
00:11:52.799 --> 00:11:57.919
So the force one was just the capability moving locally, right?
00:11:58.480 --> 00:12:11.279
Uh and this, I think we have to credit Google for pushing the edge of edge AI.
00:12:11.519 --> 00:12:20.799
That's because with their Gemma 4 model, it completely shook up uh the landscape of open source AI, right?
00:12:20.960 --> 00:12:28.080
This thing was 20 times more efficient than other open source AI models at the time.
00:12:28.159 --> 00:12:30.480
So uh essentially, let me describe it like this.
00:12:30.639 --> 00:12:32.240
Do you remember GBT 4.0?
00:12:32.559 --> 00:12:36.879
Right, one of the best models, you know, about 14-15 months ago.
00:12:37.120 --> 00:12:40.000
It was at the absolute frontier, right?
00:12:40.159 --> 00:13:01.200
So now you can download Gemma 4 on a consumer laptop, and it has, you know, roughly you look at the scientific benchmarks and the ELO score, it's essentially about a GPT-4.0 level model that you can run on your laptop.
00:13:01.360 --> 00:13:02.320
So, what's the big deal?
00:13:02.399 --> 00:13:03.759
What's the big difference, right?
00:13:04.320 --> 00:13:17.600
14-15 months ago, I mean, there's thousands of companies spending millions of dollars a year to get that type of technology, to get GPT-4.0 level technology for their employees.
00:13:17.919 --> 00:13:20.639
Now it doesn't take take anything, really.
00:13:20.799 --> 00:13:23.840
It takes a new-ish piece of Apple, right?
00:13:23.919 --> 00:13:26.159
I I just got a new MacBook Pro.
00:13:26.320 --> 00:13:30.159
That thing can run Gemma 4 very easily, right?
00:13:30.399 --> 00:13:32.720
Can run even better models than that.
00:13:32.879 --> 00:13:41.120
And and this really changes, I think, what is ultimately capable when you look at the open source versus closed source.
00:13:41.360 --> 00:13:48.399
Because yes, I think most people look at normal usage and they're comparing apples to apples, right?
00:13:48.559 --> 00:13:54.159
Here's what our marketing team did 15 months ago with a GPT-4.0 level model.
00:13:54.320 --> 00:13:57.360
Oh, now they can do that on Gemma 4.
00:13:57.519 --> 00:14:11.840
Well, yeah, you can do it, but now you can do it agentically because not only in the last year or so have the models obviously improved with now thinking models, reasoning models being the default, but now we have these agentic harnesses.
00:14:12.000 --> 00:14:20.159
Um, you know, not just the ones that you can use um, you know, inside of Chat GPT, Gemini, uh, Claude Copilot.
00:14:20.320 --> 00:14:32.960
But well, you have these local um autonomous AI systems as well, such as OpenClaw, such as uh, I always forget if it's Hermes or Hermes agent, right?
00:14:33.120 --> 00:14:51.120
Um, so now you can have essentially the level of AI from 14 months ago running for free 24-7 agentically, even if you're just doing it, you know, summarization, content creation, um, research, things like that.
00:14:51.279 --> 00:15:05.519
So when capabilities went local, right, Gemma Ford leading the way, but obviously all the Chinese models followed suit because of uh right distillation, which we'll talk about a little bit here uh in a couple of minutes.
00:15:06.000 --> 00:15:13.840
But you can't overlook Gemma because it put frontier capability on literally a laptop, right?
00:15:14.159 --> 00:15:27.759
Because two years ago, to be able to run something like a GPT 4.0 level model, uh, right, which uh rumors have been swirling that's it's a two trillion parameter model, right?
00:15:28.080 --> 00:15:34.639
You would need a small little data center to run something like that, you know, two-ish years ago.
00:15:34.799 --> 00:15:36.960
Uh now you have these capabilities.
00:15:37.200 --> 00:15:41.919
So um, you know, I've been lucky to talk to a lot of smart people in AI.
00:15:42.000 --> 00:15:49.200
And now you really have executives grappling with, well, should we be buying a bunch of, as an example, new MacBooks?
00:15:49.440 --> 00:15:59.679
Should we be, you know, buying a bunch of DGX Sparks uh for our employees and setting them up with 24-7 always on agentic AI, right?
00:15:59.759 --> 00:16:06.320
To take advantage of these now local and powerful models that, well, you don't pay, right?
00:16:06.559 --> 00:16:11.919
You download them once, you don't pay again, and they work and they can work while you sleep.
00:16:12.080 --> 00:16:19.759
Like I said, because of uh some of the new autonomous capabilities uh from local agents that can run around the clock.
00:16:20.799 --> 00:16:30.799
So this has obviously led to, and I think uh Google putting the pressure on the uh the open source world with Gemma 4.
00:16:30.879 --> 00:16:37.360
Like I said, it was 20 times more efficient in terms of what is it was able to achieve on the benchmarks in terms of size, right?
00:16:37.519 --> 00:16:45.679
Because uh when it came out, if you looked at the other Chinese models, it was about 10 to 20 times smaller in size, right?
00:16:45.840 --> 00:16:59.200
So uh you wouldn't have been able to, you know, use the best open source model pre-Gemma 4 on a local machine, uh, right, at least a consumer laptop that you can just go walk into the store and buy.
00:16:59.360 --> 00:17:00.240
Now you can.
00:17:00.480 --> 00:17:11.039
And open Chinese models are amazing, and I think they've been getting better and better and smaller and smaller and more and more efficient since Google's Gemma 4.
00:17:11.200 --> 00:17:13.920
But you have to talk about the elephant in the room.
00:17:14.160 --> 00:17:18.000
That is, these models are distilled, right?
00:17:18.319 --> 00:17:27.119
We can say that all the big labs have said that, you know, are I don't know, maybe some of our uh audience in China won't appreciate hearing that.
00:17:27.279 --> 00:17:34.400
But I mean, Google, Anthropic, OpenAI have all accused China and have said they have proof, uh, right.
00:17:34.640 --> 00:17:36.880
But the uh the White House.
00:17:36.960 --> 00:17:52.480
So in April, the White House actually officially said that China was uh using uh kind of illegal uh tactics to distill uh US AI models to create cheaper domestic knockoffs, uh right.
00:17:52.559 --> 00:17:58.799
So it got to the point that at least the White House said that they had enough uh information or intel to make that declaration.
00:17:58.960 --> 00:18:02.720
So, what is model distillation and well, why does it matter?
00:18:03.279 --> 00:18:09.920
So the easiest way is like I can spend 10 hours studying for a test, right?
00:18:10.960 --> 00:18:12.319
Think back in the classroom.
00:18:12.400 --> 00:18:14.319
I can spend 10 hours studying for a test.
00:18:15.039 --> 00:18:20.559
Someone behind me can look over my shoulder and spend 10 minutes and get the exact same answers.
00:18:20.880 --> 00:18:23.440
That's kind of like what model distillation is, right?
00:18:23.599 --> 00:18:33.519
You have the big AI companies here in the US spending billions of dollars, right, on any single new, you know, model pre-training as an example.
00:18:33.680 --> 00:18:42.400
And essentially, uh, you have uh certain actors in China who will use the API.
00:18:42.480 --> 00:18:52.799
And you know, different companies have come out with different levels of um proof and say, okay, well, they're creating you know thousands of spoof accounts, more or less.
00:18:53.039 --> 00:18:59.359
They're putting in all these inputs and training it on our models' outputs versus training it themselves.
00:18:59.519 --> 00:19:01.519
So, yeah, just kind of copying the homework.
00:19:01.680 --> 00:19:14.480
So, what this has led to is China has been able to put out these open source models, uh, really technically just pushing the frontier of open source by allegedly just copying the best US models out there.
00:19:14.559 --> 00:19:20.079
And what this has led to is well, it's a crashing out at the bottom price of intelligence.
00:19:20.240 --> 00:19:30.640
So, Deep Seek V4, as an example, DeepSeek, one of those companies that many uh of the AI labs here in the US have um accused of model distillation.
00:19:30.880 --> 00:19:42.400
DeepSeq V4 Pro, one of their newer models, now lists their price at 43 cents uh per million token inputs and 87 cents per million token outputs.
00:19:42.559 --> 00:19:50.720
That's like more than 25 times cheaper than the premium uh you know uh closed source proprietary models.
00:19:50.799 --> 00:19:56.079
And that is the reality that a lot of boardrooms are looking at right now, right?
00:19:56.240 --> 00:20:16.480
To make that math easy, it's like, okay, if we're spending a thousand dollars per month per employee on the API side, uh, right, if you're sorry, if you're spending um, let's just say$40,000 a year, okay, we could be spending a thousand dollars a year if we switch over uh to an open source model as an example.
00:20:17.279 --> 00:20:24.000
So that's here's what that actual leads that what that actually leads to, right?
00:20:24.079 --> 00:20:32.079
Kind of the uh the model distillation leads to more powerful, cheaper open source models from China.
00:20:32.480 --> 00:20:39.200
And well, it leads to people using them, but not always knowing the ramifications, right?
00:20:39.279 --> 00:20:49.920
So obviously, uh Google doing things the right way, but I think with these Chinese models, they've become increasingly popular, even in the enterprise, uh, which is tricky.
00:20:50.079 --> 00:20:57.440
And I don't think that most executives are fully understanding some of the consequences of using open source models.
00:20:57.599 --> 00:21:09.359
But this has led to essentially having a workforce of always on assistance, and they've shifted from expensive special projects to well, that's just now the default operating model.
00:21:09.599 --> 00:21:19.680
So, you know, this has just allowed kind of these the this new swarm of agentic AI that couldn't have really have existed before.
00:21:19.839 --> 00:21:31.680
Number one, the technology and the harnessing wasn't there, but number two, you take out uh, you know, at least Gemma and the Chinese models uh that it have been accused of distillation.
00:21:31.839 --> 00:21:36.720
And your your your options, aside from those, aren't really that good.
00:21:36.880 --> 00:21:42.799
All right, we're gonna talk more here after we take a quick break for a word from our partners.
00:21:43.119 --> 00:21:51.039
Adobe just introduced an entirely new way to create, bringing the power and precision of its creative suite into one conversational experience.
00:21:51.279 --> 00:21:57.440
Meet Firefly AI Assistant, now live in the Adobe Firefly app, the all in one creative AI studio.
00:21:57.519 --> 00:21:59.680
Powered by Adobe's creative agent, Firefly.
00:22:00.079 --> 00:22:07.920
Apply AI Assistant lets you start with your vision, just describe what you want, and shape the outcome as it takes form with the assistant.
00:22:08.079 --> 00:22:21.759
The Assistant orchestrates multi-step workflows, drawing on 60 plus pro-grade tools across Adobe Creative Cloud apps, including Photoshop, Illustrator Premiere, Lightroom, Express, and more to help bring your ideas to life.
00:22:22.000 --> 00:22:34.559
You can also get started with Creative Skills, a growing library of pre-built workflows for common creative tasks like batch editing photos, creating mood boards, portrait retouching, and creating social variations.
00:22:34.720 --> 00:22:40.480
Every step the assistant takes is visible, so you can refine, redirect, or take over at any time.
00:22:40.720 --> 00:22:43.279
You stay in the driver's seat as the creative director.
00:22:43.519 --> 00:22:47.119
Adobe Firefly AI Assistant, now in public beta.
00:22:47.279 --> 00:22:50.319
See it today at firefly.adobe.com.
00:22:52.480 --> 00:23:08.640
Aside from always on AI agents, what this open source movement has led to is well, now enterprises may be moving away from having the one model fits all uh solution.
00:23:08.960 --> 00:23:22.000
So now, as an example, you might be able to put out a 100-agent swarm goes from you know$1,200 plus dollars on Opus to well, maybe like$60 some dollars on Deep Seek.
00:23:22.480 --> 00:23:34.880
And now essentially you um can look at AI as more of a triage or categorical categorization of which models to use for which tasks, right?
00:23:35.039 --> 00:23:38.400
Especially when you're talking about high volume operations.
00:23:38.799 --> 00:23:48.880
So things like when you're going through it in bulk, things like summarization, uh extraction, uh parsing PDFs, classification, right?
00:23:49.279 --> 00:24:00.000
Now so many even large enterprise companies are no longer doing that on the back end using the frontier US uh companies.
00:24:00.160 --> 00:24:10.799
Well, I mean, many still are, but you've already seen a big segment uh of those companies move to these open source or uh Chinese open source models.
00:24:11.359 --> 00:24:17.119
But if you're thinking right now, if you're like, wow, our bill's pretty high, our API bill, right?
00:24:17.200 --> 00:24:24.480
I'm not talking about on the front end, you know, the number of seats you have in Chat GPT Enterprise or you know in Gemini Enterprise or anything like that, right?
00:24:24.559 --> 00:24:30.079
I'm talking about back end, uh, all of these special projects that you have running via the API.
00:24:30.240 --> 00:24:46.160
So if if you're looking at your API bill and you're like, yeah, we're going through a lot, uh, or you know, hey, we're using you know Opus 4.7 and GPT-5.5 to run our agents, maybe we should be looking at you know Kimmy or Deep Seek or uh whatever it is.
00:24:46.319 --> 00:24:52.000
Before you do that, you have to know that there is a big trade-off.
00:24:52.640 --> 00:25:02.720
Because just because a model is free or open source uh or cheap-ish to run via the API, right?
00:25:02.799 --> 00:25:10.480
If you are uh you know running some of these open models uh via the API, there's still uh an expensive price to pay.
00:25:10.720 --> 00:25:25.519
And that price might be unknown at this point, but it can cost your company a lot more than maybe just using that closed proprietary AI would have costed you could have cost you via the API.
00:25:26.559 --> 00:25:36.240
That's because using open source strips away all of that legal protection that you probably overlook or take for granted.
00:25:36.720 --> 00:25:37.680
What do I mean?
00:25:37.839 --> 00:25:44.319
Well, when you're using anthropic, open AI, Microsoft, um, who did I forget?
00:25:44.480 --> 00:25:46.559
Microsoft, open AI, Google, Anthropic, right?
00:25:46.640 --> 00:25:53.200
When you're using those uh enterprise offerings, you have a level of legal protection, right?
00:25:53.279 --> 00:25:58.559
So as an example, if you use, right, I'm not gonna go through all the uh fine prints, right?
00:25:58.640 --> 00:26:05.759
But the four companies at the enterprise level all offer, you know, some sort of essential uh uh I won't call it insurance, right?
00:26:05.920 --> 00:26:07.519
But think of it kind of like that, right?
00:26:07.599 --> 00:26:18.240
Like, hey, if you use something produced by our systems and if you use it ethically and responsibly and with guardrails and it produces something that's not you know correct, there is some level of protection there, right?
00:26:18.480 --> 00:26:22.160
Which you don't get that with open source models, right?
00:26:22.240 --> 00:26:28.880
So, as an example, you know, Deep Seek, uh, you know, they use uh different MIT licenses, Apache 2.0.
00:26:29.119 --> 00:26:35.680
Uh yeah, essentially there's no warranty or non-infringement agreements, all right.
00:26:35.920 --> 00:26:43.519
So for regulated work and customer-facing output, you have to look at the trade-off.
00:26:43.680 --> 00:26:48.079
Yeah, you might save six, seven, who knows?
00:26:48.400 --> 00:27:01.839
Maybe eight uh figures by switching the bulk of some of your maybe agentic or bulk workloads, uh, you write, especially if you're a fortune, fortune 500, fortune 100 company.
00:27:02.000 --> 00:27:10.240
There's minimum seven, eight figures that you could, in theory, save by switching some of those heavier agency uh or you know, parsing.
00:27:10.319 --> 00:27:22.720
You know, I know parsing is a big one, shifting some of those workflows to open models, but you lose that legal protection that maybe you've had to rely on it before, maybe you haven't.
00:27:22.960 --> 00:27:26.480
But that one time that you would actually need it.
00:27:26.720 --> 00:27:34.880
And if you do switch over to open source, you have to understand those ramifications because at that point you're gonna actually be paying for it.
00:27:35.759 --> 00:27:51.440
So that gets us to the real question here as we get close to wrapping up because I don't want my takeaway here to be don't use open models, they're not safe, because that's not the takeaway.
00:27:52.000 --> 00:28:00.000
I think you need to start looking at your AI workflow like a triage, right?
00:28:00.240 --> 00:28:02.799
At least when it comes to back end tasks, right?
00:28:02.960 --> 00:28:06.720
Uh, front end, I've always been a firm believer, and I still am today.
00:28:06.880 --> 00:28:14.400
You need to pick your AI operating system of choice, whether that's co-pilot, chat GPT, Claude, Gemini on the front end.
00:28:14.480 --> 00:28:22.480
And that's where you should move, especially your non-technical people, should move the majority of their day-to-day knowledge work tasks, should be happening on the front end there.
00:28:22.640 --> 00:28:26.240
But you still have a multitude of back-end tasks.
00:28:26.559 --> 00:28:33.039
And I think you have to look at it like triaging in an emergency room, right?
00:28:34.160 --> 00:28:44.799
You wouldn't send your top neurosurgeon in when someone's having an allergic reaction to honey, right?
00:28:45.119 --> 00:28:46.079
You wouldn't do that.
00:28:46.799 --> 00:28:54.319
You would save that neurosurgeon for well, someone that needs a neurosurgeon.
00:28:55.440 --> 00:29:00.799
And I think that there's so many companies that haven't gone through the basics of this.
00:29:01.039 --> 00:29:18.240
For the most part, on the API side, they pick, well, one model and they say, All right, well, we have our AI operating system of choice, and then for everything else, as an example, we go to Sonnet 4.6 or we go to you know Gemini 3.1 Flash or whatever that model may be.
00:29:18.400 --> 00:29:20.319
And maybe that's the right model.
00:29:20.559 --> 00:29:30.000
Maybe those companies have done their due diligence and have vetted out their different use cases and have priced it out, and maybe that's the right move, but maybe it's not.
00:29:30.160 --> 00:29:37.039
Because I know from experience and talking to a lot of people, a lot of companies just choose whatever is on the cutting edge.
00:29:37.200 --> 00:29:47.039
And they say, Well, this is the best, so we're gonna pay for it because there is a push internally to use more AI, to use the best AI.
00:29:47.200 --> 00:29:50.880
We see all these new benchmarks, we want to make sure that we're taking advantage of it.
00:29:51.039 --> 00:29:59.200
Well, is that neurosurgeon gonna be able to, you know, properly diagnose the allergic reaction to someone eating honey?
00:29:59.279 --> 00:30:03.039
Well, yeah, probably, but it's gonna cost you a lot more.
00:30:03.279 --> 00:30:08.319
So you need to think about sending those high-volume, low-stakes work, right?
00:30:08.400 --> 00:30:18.880
Like summarization, research, content creation, maybe, uh, to cheaper open source models that you can either run locally uh or you know, running via the API for just cost efficiency.
00:30:19.119 --> 00:30:24.640
If there's essentially like no legal ramifications if you get something wrong, right.
00:30:24.720 --> 00:30:30.079
So if you're in a highly regulated sector, this is probably not the advice for you, uh, right.
00:30:30.240 --> 00:30:37.359
You shouldn't probably be using uh, you know, or just taking my advice on a whole lot of anything as truth.
00:30:37.440 --> 00:30:40.799
You always need to be vetting these things out for yourself, uh, right.
00:30:40.880 --> 00:30:57.680
But if if if if it is something relatively in a sector that's not highly regulated, where there's not a quote unquote a lot on the line, that's one of those instances where you need to say, can we shift some of our more expensive API workloads to an open source model?
00:30:57.839 --> 00:31:04.160
Uh, or if you need to run sensitive private workflows on self-hosted open models, that's another thing.
00:31:04.319 --> 00:31:14.000
I think that there's still, even to this day, even though I think there's plenty of reasons, you know, one thing I always ask companies when they're like, Oh, we don't, well, we don't run this through AI, right?
00:31:14.079 --> 00:31:15.200
Because it's sensitive data.
00:31:15.279 --> 00:31:17.680
And I'm like, okay, well, do you have a cloud provider?
00:31:17.759 --> 00:31:18.960
And they're like, Of course.
00:31:19.200 --> 00:31:21.839
It's like, okay, well, it's the same thing, more or less, right?
00:31:22.000 --> 00:31:25.279
As long as you take proper precautions, turn off model training and all that.
00:31:25.359 --> 00:31:31.519
It's it's more or less using the same level grade of security that you know, cloud uh uses.
00:31:31.759 --> 00:31:36.079
Anyways, there are still some things that companies won't even put on the cloud, right?
00:31:36.240 --> 00:31:52.640
Which I understand, but having these now uh extremely powerful open source models and extremely efficient open source models, now you can start running those private workloads, uh uh workloads or workflows on-prem, right?
00:31:52.799 --> 00:31:56.240
Or you know, self-hosted that you can fully control.
00:31:56.880 --> 00:32:21.759
And then you can reserve those more premium, uh, those more um uh high value, highly sensitive tasks, um, you know, that for those models that can reason and think and offer kind of that that that level of uh security and legal support that you don't get if you opt for open models instead.
00:32:22.640 --> 00:32:25.519
So, you know, have a nice little chart here.
00:32:25.599 --> 00:32:43.519
So maybe when you look at the cheap open APIs, you look at simple tasks like summarization, extraction, or classification uh for local, self-hosted open models, private workflows, great for that, running agents locally, having more control, uh, right.
00:32:43.680 --> 00:32:52.720
And then for the premium closed models, which are the ones that a lot of people are using on the front end, on the back end, you should still be using these for a lot of reasons, right?
00:32:52.880 --> 00:33:00.640
Uh, those that, well, carry a lot of business value, uh hard tasks that require reasoning.
00:33:00.799 --> 00:33:02.400
Uh, right, your final review.
00:33:02.559 --> 00:33:18.559
Maybe you do uh, you know, draft version either with a cheap API or local self-hosted open model, but or anything, you know, uh that requires customer-facing output should probably be going on that premium closed models for that level of protection.
00:33:18.640 --> 00:33:33.279
Like I said, you cannot overlook the hidden trade-off that these open licenses may disclaim warranty and non-infringements, where the enterprise offerings do usually include that I that IP uh indemnification.
00:33:33.440 --> 00:33:41.839
So for regulated work, that protection in almost all cases justifies the premium that you pay.
00:33:42.240 --> 00:33:49.519
However, as we wrap up here, let me just quickly encapsulate all of this.
00:33:50.559 --> 00:33:52.400
Local models aren't going anywhere.
00:33:52.799 --> 00:33:53.119
All right.
00:33:53.279 --> 00:34:07.839
And I actually think, especially as we uh officially welcome in the era of models that can improve themselves and create can create own versions, uh, you know, smaller versions of themselves, right?
00:34:07.920 --> 00:34:17.360
All the big companies have essentially said, you know, have hinted uh at RSI or you know, the fact that our big models make smaller versions of themselves.
00:34:17.599 --> 00:34:31.119
I think that we're gonna not only see a continued trend toward uh local open source models, I think we're gonna start seeing a lot of smaller models uh for very specific use cases.
00:34:31.280 --> 00:34:33.920
It's something I've been you know predicting now for multiple years.
00:34:34.079 --> 00:34:35.360
We started to see it slowly.
00:34:35.440 --> 00:34:41.599
I think it is going to pick up steam now that we're starting to get some hints of recursive self-improvement with these models.
00:34:41.840 --> 00:34:47.119
So your company has to be paying attention because this is a trend that is not going away.
00:34:47.360 --> 00:35:00.159
The open models are going to become more and more capable, they're gonna become faster, they're gonna become more efficient, and the options are going to start to become even greater, uh, right?
00:35:00.320 --> 00:35:13.760
Not just great general purpose open models that can run on consumer hardware like Gemma 4, uh, but small open models for very specific tasks that can be highly valuable for your company.
00:35:14.000 --> 00:35:26.480
So you have to understand uh the pros and the cons of these local models uh when you might use a cheap API and how this changes the agentic outlook for your company.
00:35:26.639 --> 00:35:30.639
So don't write them off just because you always want to use the latest and the greatest.
00:35:30.719 --> 00:35:39.920
Yes, you should do that, but don't send the neurosarch in to you know triage a basic thing happening in the waiting room.
00:35:40.159 --> 00:35:43.199
Send the right model at the right time for the right purpose.
00:35:43.360 --> 00:35:50.639
So I hope this was helpful as we recapped open source AI 101 as part of our Start Here Series.
00:35:50.719 --> 00:35:53.760
If this was helpful, number one, make sure you subscribe to the podcast.
00:35:53.840 --> 00:35:54.559
I'd appreciate that.
00:35:54.639 --> 00:35:57.280
But then make sure you go to starthere series.com.
00:35:57.440 --> 00:36:01.920
That's going to give you free access to our exclusive inner circle community.
00:36:02.079 --> 00:36:06.559
Right now, there's no other way to join except by going to startheseries.com.
00:36:06.639 --> 00:36:07.519
So make sure you do that.
00:36:07.679 --> 00:36:08.400
Thank you for tuning in.
00:36:08.480 --> 00:36:11.519
I hope to see you back tomorrow and every day for more everyday AI.
00:36:11.599 --> 00:36:12.639
Thanks, y'all.
