WEBVTT
00:00:01.120 --> 00:00:09.519
If you're a heavy AI user like me, you always have your eye on the next big model release from one of the big players.
00:00:09.679 --> 00:00:20.879
And we've actually seen plenty of action on that front the last few days since the big model makers have started to work with the federal government and get the models actually released.
00:00:21.039 --> 00:00:27.519
As an example, we have Fable 5 now for a few more days, and we're going to have GPT-5.
00:00:30.399 --> 00:00:35.119
And maybe we'll have Gemini 3.5 Pro any day now.
00:00:35.359 --> 00:00:51.600
But when I think about it, in those instances, we're actually just going from having a really great model to a model that's just a little bit better than the great model we had before, which when it comes to tracking AI releases, sure, that's helpful.
00:00:51.759 --> 00:00:57.920
But are any of these a step change or, you know, really unlocking something that we didn't have before?
00:00:58.159 --> 00:01:02.159
And even with these great models, the honest answer is probably not.
00:01:02.399 --> 00:01:18.959
But this week we actually got one of those AI features that I am beyond excited for because we're going from having honestly not a good product at all when it came to real-time uh voice AI to now something that is finally good.
00:01:19.200 --> 00:01:31.680
So we're going to be going over that, the new OpenAI Chat GPT real-time voice model that's available now, as well as a bunch of other AI updates that you might have missed this week.
00:01:31.920 --> 00:01:44.560
So on today's show, in our weekly Friday features on Thursday Roundup, you're going to learn why a new voice model may be a bigger deal than Fable 5 or GPT 5.6 soul.
00:01:44.799 --> 00:01:50.799
You're going to learn how Meta may be quietly making its AI comeback, a lot of big releases from Meta this week.
00:01:50.959 --> 00:01:58.239
And you're going to see the new image model that's pushing the once untouchable ChatGPT images too.
00:01:58.640 --> 00:01:59.840
All right, let's get into it.
00:02:00.000 --> 00:02:01.120
Welcome to Everyday AI.
00:02:01.200 --> 00:02:03.920
My name's Jordan Wilson, and we do this thing well every day.
00:02:04.079 --> 00:02:14.240
This is your daily, uh, unedited, unscripted live stream podcast and free daily newsletter helping business leaders like you and me keep up with the nonstop avalanche of AI updates.
00:02:14.400 --> 00:02:16.080
I tell you what matters, what doesn't.
00:02:16.240 --> 00:02:17.280
You take that information.
00:02:17.439 --> 00:02:20.159
All of a sudden, you're the smart AI person in your company.
00:02:20.240 --> 00:02:20.800
You're welcome.
00:02:20.960 --> 00:02:27.360
Hey, to show appreciation, do me a favor, go ahead and subscribe to the show here if you're listening on Spotify or Apple Podcasts.
00:02:27.439 --> 00:02:31.039
And make sure you go to our website at your everydayai.com.
00:02:31.120 --> 00:02:33.759
Uh, we're going to be recapping the highlights from today's show.
00:02:33.840 --> 00:02:37.280
So if you hear one of these seven AI features and you're like, wait, what was that?
00:02:37.360 --> 00:02:39.120
Yeah, it's going to be in the newsletter.
00:02:39.280 --> 00:02:40.240
So don't worry.
00:02:40.400 --> 00:02:40.719
All right.
00:02:40.800 --> 00:02:44.560
Uh, and hey, FYI, we're doing this a little different this week.
00:02:44.719 --> 00:02:48.159
Normally we do the Friday AI features show on Friday.
00:02:48.319 --> 00:02:51.439
But in our newsletter, I put out a poll uh yesterday.
00:02:51.520 --> 00:02:58.960
And well, we know that Chat GPT's new model, GBD56 Soul, is going to be released well today.
00:02:59.120 --> 00:03:02.800
So on our Friday show, we're gonna be going over that new model.
00:03:02.960 --> 00:03:05.759
So today we're doing our Thursday features.
00:03:05.919 --> 00:03:09.120
So you guys voted for it overwhelmingly for the most part.
00:03:09.280 --> 00:03:10.560
I just work for you guys.
00:03:10.639 --> 00:03:16.879
So yeah, if you are a podcast listener and not on the newsletter, you gotta get on the newsletter because you're just directing me.
00:03:16.960 --> 00:03:21.520
I'm like your human agent that just, you know, uh, you know, goes to work for you.
00:03:21.759 --> 00:03:27.120
All right, so let's get into it and go over the new features.
00:03:27.360 --> 00:03:27.919
All right.
00:03:28.159 --> 00:03:32.400
So um the big one, and I think I mean it.
00:03:32.479 --> 00:03:40.400
This one technically, in terms of unlocking something, some new AI capability for our work that we didn't have before.
00:03:40.560 --> 00:03:49.759
Uh, yes, I mean, you can make the argument Fable was such a big jump, or you know, the original like 01 and 03 models from OpenAI were a big jump.
00:03:49.840 --> 00:03:50.960
And yes, they were.
00:03:51.039 --> 00:03:57.199
Uh, and yes, people are excited to have Fable 5 back for a couple more days under subscription plans.
00:03:57.280 --> 00:04:00.479
And, you know, we're gonna be getting GPT 5.6 sold any minute here.
00:04:00.719 --> 00:04:14.319
But I think the new GPT live voice model may actually be one of the biggest steps forward uh in day-to-day AI use that we've had in a very long time.
00:04:14.479 --> 00:04:16.720
That's I think for a couple of reasons.
00:04:16.800 --> 00:04:20.560
Uh, one, the models before were just really bad, right?
00:04:20.800 --> 00:04:24.160
Chat GPT's original voice mode, when it came out, it's like, oh, kind of cool.
00:04:24.240 --> 00:04:28.639
And then you had your advanced voice mode for the last like two years, and it's just really not good, right?
00:04:28.720 --> 00:04:30.959
Gemini Live, not really good.
00:04:31.120 --> 00:04:36.000
Uh, you know, Claude has a voice mode, not really good, at least, especially from the big players.
00:04:36.079 --> 00:04:47.519
There just hasn't been a great uh voice, you know, AI voice assistant that you can talk to that can pull real-time information from the web that can work with the conversation and the context that you give it.
00:04:47.680 --> 00:04:50.240
We just haven't had that, which is crazy, right?
00:04:50.319 --> 00:04:54.000
Because people are getting more and more comfortable dictating versus typing.
00:04:54.160 --> 00:04:57.360
Uh, so now this one exciting GPT Live.
00:04:57.519 --> 00:04:58.480
It is out now.
00:04:58.639 --> 00:04:59.199
It's new.
00:04:59.360 --> 00:05:00.000
Let's talk.
00:05:00.160 --> 00:05:12.399
So here's what's new OpenAI killed the kind of turn-based, you know, kind of walkie-talkie voice mode and replaced it with the new GPT Live, which is a duplex model, what they're calling it.
00:05:12.560 --> 00:05:14.800
It listens and it talks at the same time.
00:05:14.959 --> 00:05:17.519
So it also does this like back channeling like a human.
00:05:17.600 --> 00:05:21.920
So as you're talking, you know, it'll say things like, mm-hmm, got it, right?
00:05:22.000 --> 00:05:33.600
Uh, and then it redecides several times per second whether to speak, keep listening, pause, interrupt, or call a tool, like to, you know, look in the context of your uh conversation, to look on the web, etc.
00:05:34.399 --> 00:05:41.120
Also, it pulls in GPT 5.5 at the same time for web search and heavier reasoning.
00:05:41.279 --> 00:05:55.279
So, before, obviously, the voice models, you know, what they worked on, I think over the last like, you know, 2023 to 2025 was just making the voices sound more realistic and a little less latency, which is like that's cool and all.
00:05:55.439 --> 00:06:02.079
But if they don't know left from right, if they don't know what day it is, if they can't look up real-time information, what good is it?
00:06:02.240 --> 00:06:05.759
So that's what's different with this new model from OpenAI.
00:06:05.920 --> 00:06:09.199
Uh, so here is from OpenAI's blog post what they said.
00:06:09.360 --> 00:06:17.920
They said, We're launching GPT Live, a new generation of voice models that make talking with AI feel much more like having a real conversation.
00:06:18.240 --> 00:06:23.759
GPT Live is built on a full duplex architecture, meaning it can listen and speak at the same time.
00:06:23.920 --> 00:06:33.600
During conversations, GPT can show it's paying attention with phrases like, yeah, uh, engage in a quick back and forth, or just stay quiet when you need a moment to think.
00:06:33.839 --> 00:06:37.600
The result is a voice experience that is refreshingly easy to talk to.
00:06:37.759 --> 00:06:41.120
Uh, GPT Live is also our smartest voice model yet.
00:06:41.279 --> 00:06:52.160
For questions that require web search, deeper reasoning, or more complex work, it delegates to our latest frontier model behind the scenes and brings the result back into the conversation when it's ready.
00:06:52.240 --> 00:06:57.680
While it works, GPT Live can keep talking with you and maintain the flow of the conversation.
00:06:57.759 --> 00:07:08.720
Uh, at launch, GPT Live will uh use GPT 5.5 in the background as we release new frontier models, will continuously update the models used by GPT Live.
00:07:09.360 --> 00:07:12.079
So, like I said, huge news.
00:07:12.240 --> 00:07:15.040
My gosh, I've been waiting for this because I've tried.
00:07:15.279 --> 00:07:17.519
I've tried to find actual business, right?
00:07:17.600 --> 00:07:28.800
Whether I'm taking a walk or you know out in the car, or if I just like want to take a break from looking at a screen, I I've tried all every single you know voice model from all the big players.
00:07:28.879 --> 00:07:31.759
I think when Google's Gemini Live came out, it was okay.
00:07:31.920 --> 00:07:38.079
But Gemini Live actually does a bad job, strangely enough, at querying information from Google.
00:07:38.240 --> 00:07:38.800
Uh, right.
00:07:38.959 --> 00:07:39.920
So I am excited.
00:07:40.000 --> 00:07:43.920
I've gotten to play with this one a little bit, but it is brand spanking new.
00:07:44.000 --> 00:07:45.040
So who has access?
00:07:45.199 --> 00:07:46.079
Well, everyone.
00:07:46.240 --> 00:07:51.040
So if you are on the free plan, even on Chat GPT, you do have access to this.
00:07:51.199 --> 00:07:57.360
It does use a uh technically the mini version of GPT Live One, but even free users.
00:07:57.519 --> 00:08:00.800
So a billion people, yeah, a billion.
00:08:00.959 --> 00:08:04.639
That is how many weekly active users OpenAI has inside of Chat GPT.
00:08:04.720 --> 00:08:12.000
So a billion people just got a brand new, huge upgrade in what you can do with AI, right?
00:08:12.079 --> 00:08:15.839
And then on the paid tiers, you do get the full GPT Live One model.
00:08:15.920 --> 00:08:17.199
So it is live now.
00:08:17.360 --> 00:08:23.040
Make sure your app is updated, and then you just go and tap the voice button and you're off to the races.
00:08:23.199 --> 00:08:24.319
So, why is this useful?
00:08:24.480 --> 00:08:30.319
Well, it removes the dead air that made the old voice mode feel robotic, right?
00:08:30.480 --> 00:08:37.200
Uh, it you would always kind of you know interrupt each other, you're waiting, it cuts in when you don't want it to, right?
00:08:37.279 --> 00:08:40.159
It was just all voice models were really, really bad.
00:08:40.320 --> 00:08:41.840
Now, not the case anymore.
00:08:41.919 --> 00:08:49.200
And this is genuinely usable, hands-free for brainstorming, tutoring, and live QA where interruption actually matters.
00:08:49.360 --> 00:08:50.639
So, who's gonna find it valuable?
00:08:50.799 --> 00:08:56.080
Well, anyone using ChatGPT while you're driving, walking, cooking, uh, you know, packing for a trip.
00:08:56.240 --> 00:09:04.879
I think there's a lot of you know, cool use cases uh that with it being able to reason and look things up in real time that we just didn't have before.
00:09:05.120 --> 00:09:06.320
Um, right.
00:09:06.480 --> 00:09:17.519
And we've seen rumors over the last year or so that you know open AI was working on some hardware devices, those things, you know, allegedly got tabled, the Johnny Ives, you know, uh I.O.
00:09:17.759 --> 00:09:18.320
acquisition.
00:09:18.399 --> 00:09:24.240
So maybe this is the model that will power future uh OpenAI hardware when and if that comes.
00:09:24.480 --> 00:09:26.559
If so, that makes a lot of sense.
00:09:26.639 --> 00:09:33.759
Uh, because I think right now Google Home uh with Gemini is the only one that has a true smart AI home speaker.
00:09:33.840 --> 00:09:41.840
Uh, and even the difference now uh between GBT Live and Gemini Home, I mean GBT Live is in another field, at least for now.
00:09:41.919 --> 00:09:43.360
We'll see what the rest of the labs do.
00:09:43.440 --> 00:09:44.960
So uh pretty big news there.
00:09:45.120 --> 00:09:48.960
Let's go on to our second update, which is technically related.
00:09:49.039 --> 00:09:53.200
Uh, so this is kind of the developer version of this model.
00:09:53.519 --> 00:10:03.039
So from OpenAI, they also released uh the day before, so uh they released this on Tuesday two new API voice models.
00:10:03.200 --> 00:10:09.679
These are called GPT Real Time 2.1, and then the budget-friendly GPT Real Time 2.1 Mini.
00:10:10.080 --> 00:10:18.879
So this allows you to add adjustable reasoning, depth, native tool calling, and better handling of alpha numerics, um, right?
00:10:19.039 --> 00:10:24.159
Things like you know, ordering numbers, codes, etc., silence, noise, and interruptions.
00:10:24.240 --> 00:10:29.120
Uh, so open AI did say they also cut latency in this voice model by 25%.
00:10:29.679 --> 00:10:32.240
Um, and this is for developers, right?
00:10:32.320 --> 00:10:41.519
So if you uh are using the back end of the open AI platform, uh this was launched, like I said, uh this week and it is generally available.
00:10:41.600 --> 00:10:46.559
You can use it in the um open AI playground, um, but this isn't like a consumer feature.
00:10:46.639 --> 00:10:49.279
So this isn't what you're gonna, you know, log into Chat GPT.
00:10:49.360 --> 00:10:53.039
In that aspect, you're using the new GPT live model.
00:10:53.200 --> 00:10:58.879
Uh for um builders, you are using the GPT real-time now 2.1.
00:10:59.039 --> 00:11:03.279
Uh, so uh who's gonna, you know, why is this useful?
00:11:03.360 --> 00:11:04.399
Who's gonna find it valuable?
00:11:04.559 --> 00:11:13.440
I think developers, product teams who are building voice agents, uh, right, call center automation, uh, anyone working on a cost-sensitive voice startup.
00:11:13.600 --> 00:11:23.440
So even if you're not a true developer, right, this is something I'm gonna be using uh because I've wanted to build something, it probably doesn't take long.
00:11:23.600 --> 00:11:25.360
Maybe I'll do it as a live show once.
00:11:25.519 --> 00:11:31.120
It's just hard to sometimes use my phone and audio when I'm trying to do things that require audio.
00:11:31.279 --> 00:11:39.279
Anyways, um, you know, I've wanted to build a simple uh computer control agent that I can talk to and say, hey, open this, open that, right?
00:11:39.360 --> 00:11:52.240
So kind of like what I do in Codex, I dictate to Codex, but I thought, hey, it'd be cool if I just had an assistant that lives and it's always on and always listening, and I can just talk to it and it can do anything that you know the uh open AI models can do.
00:11:52.320 --> 00:11:57.600
So that's an example of something that you could build, and you can do it fairly easily just you know using codex.
00:11:57.840 --> 00:12:00.799
Uh so developers, product teams are gonna find it useful.
00:12:00.879 --> 00:12:14.159
But I mean, why it's actually useful is well, they just the lower latency and then the built-in tools mean that any voice agents that you or your team build will actually feel real time instead of laggy.
00:12:14.799 --> 00:12:16.159
All right, let's go on.
00:12:16.320 --> 00:12:18.639
Our third one, we have a new model.
00:12:18.879 --> 00:12:20.639
Is it in the frontier?
00:12:20.879 --> 00:12:22.080
Maybe we'll see.
00:12:22.240 --> 00:12:30.320
We don't have a lot of benchmarks yet, but let's talk about what we do know about Grok 4.5, the new model hot off the presses.
00:12:30.399 --> 00:12:34.879
So this is XAI, or are they called Space XAI now?
00:12:34.960 --> 00:12:35.759
I I don't even know.
00:12:35.840 --> 00:12:37.600
I think they're just XAI, right?
00:12:37.679 --> 00:12:45.360
So this is their newest flagship model, uh, tuned hard for software engineering and multi-step agentic work.
00:12:45.519 --> 00:12:54.000
Uh, so the headline claim is this XAI says that it uses four times fewer tokens than comparable models on average.
00:12:54.159 --> 00:13:00.559
So, although this isn't something that's got that's benchmarking off the charts, it is somewhat comparable.
00:13:00.720 --> 00:13:11.919
Um, you know, as an example, I'm looking at their announcement post here um on you know deep suite one one, which is I think one of the best uh software engineering benchmarks.
00:13:12.080 --> 00:13:22.720
Uh, you know, it's not great, but it's you know, at least in the tier with you know, Fable, GPT-5.5, and Opus 4.8.
00:13:22.799 --> 00:13:25.759
Uh, same thing with terminal bench, it's much more competitive.
00:13:25.840 --> 00:13:32.320
It's actually ahead of Opus 4.8, uh, and pretty close there to GPT-5.5 and Fable as well.
00:13:32.480 --> 00:13:40.000
Um, and then Sui Brand Pro, same thing, a little ahead of the um Opus models and uh GPT-5.5 there.
00:13:40.159 --> 00:13:45.840
So great uh model, specifically, I think, for software engineering.
00:13:46.000 --> 00:13:48.159
So here is what Grok says.
00:13:48.320 --> 00:13:51.039
They said today we're launching Grok 4.5.
00:13:51.200 --> 00:13:51.919
Yeah, there we go.
00:13:52.080 --> 00:13:56.639
Space XAI's see, is this XAI or Space XAI?
00:13:56.720 --> 00:13:58.240
They say it both ways, anyways.
00:13:58.399 --> 00:14:05.600
Uh, this is uh Space XAI's smartest model built to excel at coding, agentic tasks, and knowledge work.
00:14:05.679 --> 00:14:09.679
It is our strongest model ever and was trained alongside Cursor.
00:14:10.000 --> 00:14:11.519
Uh, more on that here in a second.
00:14:11.600 --> 00:14:14.080
So they say real world engineering excellence.
00:14:14.240 --> 00:14:21.919
Groc 4.5 was trained on data sets spanning knowledge and coding, science engineering, and math with both intelligent and efficient reasoning.
00:14:22.159 --> 00:14:29.360
Groc 4.5 excels at real engineering tasks and exceeds comparable leading models at these tasks.
00:14:29.519 --> 00:14:32.960
Uh, so a little bit more to know.
00:14:33.279 --> 00:14:36.240
Uh, well, right now, where can you use this?
00:14:36.399 --> 00:14:42.480
So, well, you can use it in the Grok app, but the access depends on if what plan you're on.
00:14:42.639 --> 00:14:45.120
So, free users get like hardly nothing.
00:14:45.279 --> 00:14:53.279
So, uh, if you're on a paid plan, you will have some limited use of Grok 4.5, or if you're using Grok build.
00:14:53.440 --> 00:14:54.799
So, why is it useful?
00:14:54.960 --> 00:15:00.080
Uh, well, hey, we'll see how the third-party benchmarks uh look, right?
00:15:00.159 --> 00:15:02.080
I'll be looking at artificial analysis.
00:15:02.240 --> 00:15:06.240
Uh, cost to run the test is always one of the most telling benchmarks.
00:15:06.480 --> 00:15:24.480
But if this is true, if you are able, at least on the software engineering side, able to get you know, Opus, you know, five five level um output, level intelligence at you know about a fourth of the cost, that's pretty big, right?
00:15:24.639 --> 00:15:38.240
Uh, so this is probably not gonna be a model that you know your whole team jumps in on and you're using it for all kinds of things, but this could uh in the future aim to be a specialist dev software engineering model, especially with cursor, right?
00:15:38.320 --> 00:15:44.879
So SpaceX uh acquired uh cursor as part of you know XAI, SpaceX going public.
00:15:45.200 --> 00:15:49.279
And the thing that they have that's a huge advantage is they have compute, right?
00:15:49.440 --> 00:15:51.679
Them and Meta have tons of compute.
00:15:51.759 --> 00:15:57.120
They're one of the probably the only you know, quote unquote AI companies that just have compute to spare.
00:15:57.200 --> 00:15:59.840
That's why they're running it out and they've become kind of a neo-cloud.
00:16:00.159 --> 00:16:06.159
Um, but this and that combined with cursor is going to be a big deal.
00:16:06.240 --> 00:16:11.200
So cursor working with um the XAI or Space XAI team.
00:16:11.360 --> 00:16:19.840
Uh, I'm sure that Grok and um Cursor's model, Composer uh 2.5, are gonna start to intertwine and you know learn from each other.
00:16:19.919 --> 00:16:25.840
But I I would probably be more interested in Composer 3 uh than Grok 4.5.
00:16:25.919 --> 00:16:32.960
But regardless, uh, for a lot of people, if you're out there building, you do at least have a new model to take a look at.
00:16:33.200 --> 00:16:38.639
All right, we have a lot more to get to, but first, quick break for a word from our partners.
00:16:39.120 --> 00:16:44.159
Here's what most AI tools still can do: work outside their own little box.
00:16:44.320 --> 00:16:46.639
The all new Slack bot just changed that.
00:16:46.799 --> 00:16:53.679
It's your AI teammate inside Slack, and now it can read, write, and act across the other apps your team already uses.
00:16:53.840 --> 00:16:59.200
No more tool switching, no more re-explaining yourself every time you open a new tab.
00:16:59.360 --> 00:17:05.759
One ops team at engine says the summary feature alone saves them 15 to 20 minutes of use.
00:17:05.920 --> 00:17:09.279
See what Slackbot can do at Slack.com.
00:17:12.160 --> 00:17:16.960
All right, uh, let's keep it going for our Friday features on Thursday.
00:17:17.119 --> 00:17:24.400
Yeah, during during that ad read, I don't know if you've ever done this where you have an old coffee, but you didn't know it was old, and then I almost spit it out.
00:17:24.640 --> 00:17:26.960
Um, yeah, that that'll wake you up.
00:17:27.200 --> 00:17:33.359
All right, uh our next AI feature, Notion Agents, a dedicated iPhone app.
00:17:33.680 --> 00:17:38.319
So here's what's new, why it matters, and who's gonna find it valuable.
00:17:38.480 --> 00:17:46.720
So this is a brand new standalone iPhone app from Notion, which is separate from the main Notion app.
00:17:46.880 --> 00:17:52.240
And this is out now, it is released and it's built purely for running Notion's AI agents.
00:17:52.480 --> 00:18:06.079
So you can capture by text, voice note, or photo, and then answers uh agents answer from your docs and connected tools and can take actions like creating pages, drafting updates, and running team workflows.
00:18:06.160 --> 00:18:12.319
And this is obviously model flexible, so it can work with Claude, Gemini, ChatGPT, whatever your team uses.
00:18:12.480 --> 00:18:13.759
So who has access?
00:18:14.000 --> 00:18:16.799
Well, it is live on the app store.
00:18:16.880 --> 00:18:24.880
So it is iPhone first right now, and it does run on iPad and Mac, but isn't really designed for them and no Android announcement yet.
00:18:25.359 --> 00:18:28.000
Uh so no pricing, dedicated pricing.
00:18:28.079 --> 00:18:34.960
But if you do have a paid, uh, you know, your company has a paid um notion plan, you're you're in, right?
00:18:35.039 --> 00:18:44.559
So uh it does notions, agent in AI features for the most part do require a paid plan, and those go anywhere from you know about$10 to$20 per user per month.
00:18:44.720 --> 00:18:48.160
Um, so it does tie into your existing paid plan.
00:18:48.319 --> 00:18:49.359
So why is it useful?
00:18:49.519 --> 00:18:56.480
Well, it lets you fire off workspace tasks and capture ideas from your phone instead of waiting to get to a desk.
00:18:56.559 --> 00:19:00.960
Uh, and then the voice and the photo capture kind of kill the friction of logging things on mobile.
00:19:01.119 --> 00:19:02.720
So who's gonna find this valuable?
00:19:02.880 --> 00:19:10.240
Well, if you are a team that uses Notion, uh and Notion has been, I think, crushing it on the AI side, right?
00:19:10.400 --> 00:19:17.680
I think Notion and Slack are two companies that, you know, a year and a half ago, if you would have said, hey, would they be in this kind of like second tier?
00:19:17.839 --> 00:19:26.799
I would have said, probably not, but sure enough, you know, I think Notion, Slack, and some others are, you know, really creeping up in terms of capabilities and what they can do.
00:19:26.960 --> 00:19:35.920
So, you know, if your team is on Notion and you haven't started yet taking advantage uh of their agentic features, they're actually really, really good, right?
00:19:36.000 --> 00:19:48.559
And uh met with a couple of weeks ago when I was out in San Francisco, I actually uh met with the Notion team out there, and you know, they're they're working on some uh pretty capable uh offerings, so it is something to take uh to keep an eye on.
00:19:48.799 --> 00:19:53.440
Uh so uh although I I I'm not super sure, right?
00:19:53.519 --> 00:19:56.240
I'm not a uh power notion user myself.
00:19:56.400 --> 00:19:58.799
I've used it you know off and on throughout the years.
00:19:58.960 --> 00:20:01.440
I'm not certain about the separate app.
00:20:01.680 --> 00:20:09.759
Um, so I'm not sure why it's not just built into their normal app and then having a uh an agents section.
00:20:09.920 --> 00:20:14.880
I'm sure there's some you know uh technical reason why.
00:20:15.039 --> 00:20:21.519
Um, but we'll see uh if that sticks, if it gets popular, if it doesn't, etc.
00:20:21.839 --> 00:20:22.960
All right, next.
00:20:23.359 --> 00:20:30.160
This one again, this might be for a smaller audience, but it's something that I think could make a big deal.
00:20:30.319 --> 00:20:32.160
So Google Voice AI.
00:20:32.319 --> 00:20:44.400
Yeah, so I think for a lot of small business, small business owners, if you don't know Google Voice, it is a way to just get a uh phone number, a phone line for your company in like two minutes and for super cheap.
00:20:44.559 --> 00:20:49.039
And now they're rolling out a ton of new, very helpful AI features.
00:20:49.200 --> 00:20:54.319
Uh so Google Voice has added Gemini powered take notes for me, right?
00:20:54.400 --> 00:20:59.279
So you can just tap notes mid-call and it records and transcribes everything automatically.
00:20:59.359 --> 00:21:06.000
And then after the call, you get a summary with key points and action items plus the full transcript and audio.
00:21:06.160 --> 00:21:08.720
And then the notes arrive as an email link.
00:21:08.880 --> 00:21:14.400
So here is what Google uh said about this new uh feature.
00:21:15.039 --> 00:21:21.200
So uh they say Google Voice launched over 15 years ago as a call forwarding and texting service within the US.
00:21:21.359 --> 00:21:27.519
Since then, it has evolved into a trusted global communications platform for users and businesses alike.
00:21:27.680 --> 00:21:29.680
We're taking the next step in that evolution.
00:21:29.839 --> 00:21:41.200
We're bringing the power of AI note-taking directly to your phone calls and introducing new plans designed to help users scale their businesses with a professional grade phone system at an industry leading price.
00:21:41.440 --> 00:21:46.240
So, this does require a paid Google Voice standard or premiere plan.
00:21:46.400 --> 00:21:49.359
So the free starter tier does not include it.
00:21:49.519 --> 00:21:52.880
Uh, right now it's English only for users in the US.
00:21:53.039 --> 00:21:55.119
I'm sure that will change here soon.
00:21:55.359 --> 00:21:55.839
All right.
00:21:56.000 --> 00:21:58.960
So, um, well, why is it useful?
00:21:59.119 --> 00:22:05.200
So if you Which I do know a lot of small businesses that use uh Google Voice, right?
00:22:05.279 --> 00:22:16.640
So I guess the downside is well, you they it's been behind some of the other you know, VoIP uh providers uh out there that have had some AI features uh before them.
00:22:16.880 --> 00:22:26.240
So not saying this is going to be a differentiator for businesses, but especially for small businesses that are out in the field, they have physical locations.
00:22:26.400 --> 00:22:27.920
This one's a big one.
00:22:28.079 --> 00:22:37.759
So, you know, uh it is able to, you know, turn all those phone calls into searchable records and you know, ready-made follow-up tasks with zero manual note-taking.
00:22:37.839 --> 00:22:50.960
Uh, so the standalone pricing, also, you know, if you're a solo operator, if you have a side hustle, small business, I think it's uh a way to cheaply uh bring some AI, especially if you're uh a type of business that's on the phone all the time.
00:22:51.200 --> 00:22:51.839
All right.
00:22:52.480 --> 00:22:54.799
Our next AI update.
00:22:55.039 --> 00:23:02.880
Yeah, uh Chat GPT Images, their new images 2.0 model that's been absolutely dominating since it came out.
00:23:03.039 --> 00:23:16.480
No one has been able to touch the uh OpenAI GPT Images 2 model when it comes to all benchmarks, but they may have company near the top, not passing them yet, but at least now a more serious uh contender.
00:23:16.799 --> 00:23:22.960
So this is ByteDance, their new model, CDream 5.0 Pro.
00:23:23.920 --> 00:23:29.359
So this is ByteDance's new image model built to understand design, not just generate images.
00:23:29.519 --> 00:23:36.319
So that at least from some of the samples that I've seen, uh, this is seems to be kind of where they're pushing heavy in.
00:23:36.400 --> 00:23:46.319
So adding that understanding or reasoning layer to image generation and not just having it be, you know, this diffusion technology that just decides what pixels go where.
00:23:46.480 --> 00:23:52.720
Some of the more impressive examples that I'll kind of scroll to for our live stream audience here, you know.
00:23:52.880 --> 00:24:01.119
So as an example showing like the six major types of tea, uh, right, but it has to be able to reason and think uh and break that down.
00:24:01.200 --> 00:24:06.960
There's a brewing temperature breakdown, you know, kind of a uh oxidation level.
00:24:07.119 --> 00:24:12.640
So there's a lot of reasoning and logic and thinking and planning that goes into some of these images, right?
00:24:12.799 --> 00:24:21.200
So if you think about things that might be used as an infographic for your company, you know, just strong imagery that can really support a more visual story.
00:24:21.279 --> 00:24:31.119
Uh, this, you know, is probably now in that top three, I think, you know, along with Nano Banana uh Pro and GPT Images 2.
00:24:31.279 --> 00:24:37.920
You now have Seed Dance, uh sorry, Seed Dream, not Seed Dance, Seed Dance is the video, Seed Dream is the photo version.
00:24:38.079 --> 00:24:38.480
All right.
00:24:38.559 --> 00:24:51.599
So it also adds that data and infographic visualization from dense text, interactive precision editing, uh, and stronger photorealism and uh native generation in 10 languages.
00:24:51.920 --> 00:24:57.200
But uh one of the standout features, I think, is a layer separation piece.
00:24:57.359 --> 00:25:02.480
Uh, so it edges toward it being kind of like a design tool, not just like a prompt box.
00:25:02.640 --> 00:25:14.079
So if you've ever used, you know, Canva or Photoshop and having all these different layers and being able to adjust it via layers versus as a flat image is gonna save a lot of time and give you more granular control.
00:25:14.319 --> 00:25:20.400
So this is announced and it is available inside ByteDance's own apps right now.
00:25:20.559 --> 00:25:25.599
So these are probably not apps that you use, uh, but it's in Dreamania.
00:25:26.000 --> 00:25:30.000
Uh, and then for uh Chinese users, it's in Jim Mang.
00:25:30.240 --> 00:25:36.799
Uh, so for international users, you can download the uh Dream Mania app from ByteDance.
00:25:36.960 --> 00:25:47.839
Right now, there is no public API available yet, uh, but there are some third-party platforms uh, you know, that you know are conglomerates where you can use a bunch of the different.
00:25:48.240 --> 00:25:49.759
I think fall is one of them.
00:25:49.839 --> 00:25:52.640
Is it fall or foul FAL or is it FAI?
00:25:52.720 --> 00:25:53.119
Uh right.
00:25:53.200 --> 00:25:55.440
So there's a bunch of these third-party ones that have it.
00:25:55.519 --> 00:26:06.160
So if you really do want to uh get your hands on it, if you use one of those third-party um, you know, AI image generators that have model selection, you're probably gonna see it in there.
00:26:06.720 --> 00:26:08.880
So uh why is it useful?
00:26:09.039 --> 00:26:14.319
Well, like I said, it is instantly a top three, maybe top two image model.
00:26:14.480 --> 00:26:19.279
So it turns data and dense text into clean on-brand layouts.
00:26:19.440 --> 00:26:36.400
So if for whatever reason, if you can't use Chat GPT or if you are a heavy into marketing, uh design, advertising, and you're not getting the results that you like out of Nano Banana or GPT Images 2, you now have another player to at least try.
00:26:36.559 --> 00:26:49.039
Uh so also uh multilingual teams, if you need text accurate output across different languages, designers, marketers, content teams who are producing you know infographics and branded visuals.
00:26:49.119 --> 00:26:54.400
Uh, this is definitely a new AI feature you can use now that you're gonna want to check out.
00:26:54.559 --> 00:27:01.359
All right, and then last but not least, Meta is back, back again.
00:27:01.680 --> 00:27:02.640
Meta's back.
00:27:02.799 --> 00:27:03.839
Tell some friends.
00:27:03.920 --> 00:27:04.480
All right.
00:27:04.640 --> 00:27:08.000
Uh, so let me just get this out there.
00:27:08.720 --> 00:27:14.400
Meta, when they came out with their, you know, they have their MSL, their meta superintelligence labs.
00:27:14.480 --> 00:27:19.440
They spent, you know, billions of dollars hiring all this talent, billions of dollars on compute.
00:27:19.599 --> 00:27:22.319
And then, you know, everyone thought, oh, we're gonna get a new llama.
00:27:22.400 --> 00:27:27.440
Instead, they went a different direction with the Muse, uh, that is their large language model.
00:27:27.599 --> 00:27:30.960
And there's some rumors that they're coming out with a bigger version of it.
00:27:31.039 --> 00:27:36.319
So we have Muse Spark, which is presumably uh that came out about two months ago, and that's their smaller model.
00:27:36.480 --> 00:27:43.920
And actually, at the time it came out, through a lot of benchmarks, it was maybe like a top four or five model, not anymore, right?
00:27:44.000 --> 00:27:49.920
Now it's probably in the you know eight, nine, 10 range after a lot of new updates from all the other big players.
00:27:50.160 --> 00:27:57.359
But you can't knock Meta for going from like zero to top five with their first large language model in Muse.
00:27:57.599 --> 00:28:10.160
So they've actually done the same thing with Muse image and muse video because if you're looking at Arena, uh, right, which is kind of the blind taste test for image and video models, these are top three models, right?
00:28:10.240 --> 00:28:15.440
So going from zero, never having at least under the Muse family.
00:28:15.519 --> 00:28:24.319
I know they did uh have you know their image models previously, but you know, they scrapped everything, started, you know, gotten rid of the the llama baseline.
00:28:24.559 --> 00:28:28.720
So pretty, pretty interesting results for our you know live stream audience.
00:28:28.880 --> 00:28:30.079
You can see some of these examples.
00:28:30.160 --> 00:28:31.680
We shared them in our newsletter.
00:28:31.920 --> 00:28:34.960
Pretty good, both on the photo and the video side.
00:28:35.119 --> 00:28:37.519
So here's a little bit what it is, what's new.
00:28:37.599 --> 00:28:52.480
So Muse Image is Meta Superintelligences Labs first or MSL, their first in-house image model built as an agent that can use search and code execution to self-refine its output.
00:28:52.799 --> 00:29:05.039
So, kind of the uh maybe controversial but signature feature is you can generate or manipulate images of a public Instagram account just by tagging it, right?
00:29:05.119 --> 00:29:13.920
So you kind of had this originally in Sora from OpenAI, which is no longer RAP Sora, they had this thing called cameos, right?
00:29:14.000 --> 00:29:16.319
But people had to open up, right?
00:29:16.400 --> 00:29:26.720
Like, yes, you can use my cameo, but it looks like if you're a certain level of public Instagram account, uh people can do it just by tagging you.
00:29:26.880 --> 00:29:32.960
All right, so Muse Video is built on the same foundation with native audio, but is still in development.
00:29:33.039 --> 00:29:34.799
So no release on that.
00:29:34.960 --> 00:29:49.279
Uh, so this is uh on the images side, it is free for everyday creation if you are using the meta AI app uh or meta.ai inside Instagram, WhatsApp, Facebook, Messenger, all those things.
00:29:49.440 --> 00:29:56.480
Uh, you should see uh the new um meta muse image model available.
00:29:56.559 --> 00:29:59.119
Uh, but heavy users are going to pay.
00:29:59.200 --> 00:30:06.240
So there is a kind of a free daily usage threshold, uh, and then it rolls into the new meta paid subscription plans.
00:30:06.319 --> 00:30:09.759
Uh, they didn't you know disclose the exact limit or price.
00:30:09.920 --> 00:30:13.680
Again, this is pretty new, uh, but there's no public API right now.
00:30:13.839 --> 00:30:23.200
So uh if if you're not on a paid meta plan, which I don't know anyone that's on a paid meta subscription, I know they just rolled that out a couple of months ago.
00:30:23.279 --> 00:30:32.799
Uh, you know, unless your business like runs on meta in Instagram, I don't know anyone that's actually subscribing to Meta for AI features, right?
00:30:32.880 --> 00:30:44.160
So it will be uh worth tracking when and if they come out on the API side, because at least the samples of the video model and the photo model look really good, right?
00:30:44.240 --> 00:30:51.359
But if it's fairly limited to what you can use for free, I don't know how much traction this is gonna get, right?
00:30:51.519 --> 00:31:02.079
Especially when presumably we may be getting within you know three to six months, we may be getting brand new updates to GBT images and to nano banana.
00:31:02.240 --> 00:31:10.720
So at that point, and you know, we just talked about uh the new uh you know image model uh seed dream from Bite Dance.
00:31:10.799 --> 00:31:13.680
So I don't know if I was meta, I would have came out a little more hard.
00:31:13.759 --> 00:31:18.079
Like, hey, use this thing, use it a ton because the results are pretty good.
00:31:18.160 --> 00:31:26.240
And like I said, Arena, I believe once they were released, they were both top three, top four, you know, image and video models, so fairly impressive.
00:31:26.319 --> 00:31:27.279
So, why is it useful?
00:31:27.440 --> 00:31:35.119
Well, number one, it's free right now, limited use, and it lives inside apps that billions of people already use.
00:31:35.279 --> 00:31:40.160
So, this is no new tool to learn, it's just in the apps that people use.
00:31:40.480 --> 00:31:49.039
Also, interesting to see how this plays out, but the agentic self-refinement aims for just kind of more accurate, less hallucinated images.
00:31:49.200 --> 00:31:58.960
So I think you know, the people or the groups that are gonna find this most valuable, especially early on, is if your job right now requires, right?
00:31:59.039 --> 00:32:08.559
If you're a social media manager, social media marketer, then you obviously are living inside and you know, living and breathing on your, you know, meta metrics.
00:32:08.720 --> 00:32:10.400
That's this is gonna be for you, right?
00:32:10.480 --> 00:32:19.599
So, versus having to work with an external tool, meta's obviously made this, you know, um uh this earlier technology available in their ads.
00:32:19.680 --> 00:32:21.599
I would assume that we'll see that here soon as well.
00:32:21.839 --> 00:32:35.759
So if you are a content creator, social media marketing, social media advertising, and you lay heavily uh on any of the meta, you know, Instagram, Facebook, WhatsApp platforms, this is gonna be one worth checking out.
00:32:35.920 --> 00:32:38.480
All right, so that is a wrap, y'all.
00:32:38.640 --> 00:32:40.799
Our seven big features.
00:32:40.960 --> 00:32:42.799
Let's do them in reverse order.
00:32:42.960 --> 00:33:22.880
So the new meta muse image and video models, uh, bite dances, new C Dream 5.0 Pro on the image side, Google Voice AI, great for small businesses, Notion Agents dedicated iPhone app for uh teams using uh Notion Grock 4.5 new model could be good for software engineering, open AI real time 2.1, the update to GPT Real Time 2.0, can do a lot of things agentically, and then last but not least, which I think may ultimately end up making the biggest difference for millions of business users, is the new GPT live model.
00:33:23.279 --> 00:33:26.400
So I'll probably end up doing uh a review on that.
00:33:26.559 --> 00:33:28.000
But thank you for tuning in.
00:33:28.160 --> 00:33:30.319
Thanks for the curveball again, right?
00:33:30.480 --> 00:33:34.319
Last week we had the uh July 4th observed holiday.
00:33:34.480 --> 00:33:41.359
This week we're switching it up so tomorrow we can do the show on Soul GPT 56 Soul.
00:33:41.599 --> 00:33:43.039
So make sure to join us for that.
00:33:43.200 --> 00:33:44.160
Thank you for tuning in.
00:33:44.319 --> 00:33:47.599
Hope to see you back tomorrow and every day for more everyday AI.
00:33:47.680 --> 00:33:48.640
Thanks, y'all.
00:00:01.120 --> 00:00:09.519
If you're a heavy AI user like me, you always have your eye on the next big model release from one of the big players.
00:00:09.679 --> 00:00:20.879
And we've actually seen plenty of action on that front the last few days since the big model makers have started to work with the federal government and get the models actually released.
00:00:21.039 --> 00:00:27.519
As an example, we have Fable 5 now for a few more days, and we're going to have GPT-5.
00:00:30.399 --> 00:00:35.119
And maybe we'll have Gemini 3.5 Pro any day now.
00:00:35.359 --> 00:00:51.600
But when I think about it, in those instances, we're actually just going from having a really great model to a model that's just a little bit better than the great model we had before, which when it comes to tracking AI releases, sure, that's helpful.
00:00:51.759 --> 00:00:57.920
But are any of these a step change or, you know, really unlocking something that we didn't have before?
00:00:58.159 --> 00:01:02.159
And even with these great models, the honest answer is probably not.
00:01:02.399 --> 00:01:18.959
But this week we actually got one of those AI features that I am beyond excited for because we're going from having honestly not a good product at all when it came to real-time uh voice AI to now something that is finally good.
00:01:19.200 --> 00:01:31.680
So we're going to be going over that, the new OpenAI Chat GPT real-time voice model that's available now, as well as a bunch of other AI updates that you might have missed this week.
00:01:31.920 --> 00:01:44.560
So on today's show, in our weekly Friday features on Thursday Roundup, you're going to learn why a new voice model may be a bigger deal than Fable 5 or GPT 5.6 soul.
00:01:44.799 --> 00:01:50.799
You're going to learn how Meta may be quietly making its AI comeback, a lot of big releases from Meta this week.
00:01:50.959 --> 00:01:58.239
And you're going to see the new image model that's pushing the once untouchable ChatGPT images too.
00:01:58.640 --> 00:01:59.840
All right, let's get into it.
00:02:00.000 --> 00:02:01.120
Welcome to Everyday AI.
00:02:01.200 --> 00:02:03.920
My name's Jordan Wilson, and we do this thing well every day.
00:02:04.079 --> 00:02:14.240
This is your daily, uh, unedited, unscripted live stream podcast and free daily newsletter helping business leaders like you and me keep up with the nonstop avalanche of AI updates.
00:02:14.400 --> 00:02:16.080
I tell you what matters, what doesn't.
00:02:16.240 --> 00:02:17.280
You take that information.
00:02:17.439 --> 00:02:20.159
All of a sudden, you're the smart AI person in your company.
00:02:20.240 --> 00:02:20.800
You're welcome.
00:02:20.960 --> 00:02:27.360
Hey, to show appreciation, do me a favor, go ahead and subscribe to the show here if you're listening on Spotify or Apple Podcasts.
00:02:27.439 --> 00:02:31.039
And make sure you go to our website at your everydayai.com.
00:02:31.120 --> 00:02:33.759
Uh, we're going to be recapping the highlights from today's show.
00:02:33.840 --> 00:02:37.280
So if you hear one of these seven AI features and you're like, wait, what was that?
00:02:37.360 --> 00:02:39.120
Yeah, it's going to be in the newsletter.
00:02:39.280 --> 00:02:40.240
So don't worry.
00:02:40.400 --> 00:02:40.719
All right.
00:02:40.800 --> 00:02:44.560
Uh, and hey, FYI, we're doing this a little different this week.
00:02:44.719 --> 00:02:48.159
Normally we do the Friday AI features show on Friday.
00:02:48.319 --> 00:02:51.439
But in our newsletter, I put out a poll uh yesterday.
00:02:51.520 --> 00:02:58.960
And well, we know that Chat GPT's new model, GBD56 Soul, is going to be released well today.
00:02:59.120 --> 00:03:02.800
So on our Friday show, we're gonna be going over that new model.
00:03:02.960 --> 00:03:05.759
So today we're doing our Thursday features.
00:03:05.919 --> 00:03:09.120
So you guys voted for it overwhelmingly for the most part.
00:03:09.280 --> 00:03:10.560
I just work for you guys.
00:03:10.639 --> 00:03:16.879
So yeah, if you are a podcast listener and not on the newsletter, you gotta get on the newsletter because you're just directing me.
00:03:16.960 --> 00:03:21.520
I'm like your human agent that just, you know, uh, you know, goes to work for you.
00:03:21.759 --> 00:03:27.120
All right, so let's get into it and go over the new features.
00:03:27.360 --> 00:03:27.919
All right.
00:03:28.159 --> 00:03:32.400
So um the big one, and I think I mean it.
00:03:32.479 --> 00:03:40.400
This one technically, in terms of unlocking something, some new AI capability for our work that we didn't have before.
00:03:40.560 --> 00:03:49.759
Uh, yes, I mean, you can make the argument Fable was such a big jump, or you know, the original like 01 and 03 models from OpenAI were a big jump.
00:03:49.840 --> 00:03:50.960
And yes, they were.
00:03:51.039 --> 00:03:57.199
Uh, and yes, people are excited to have Fable 5 back for a couple more days under subscription plans.
00:03:57.280 --> 00:04:00.479
And, you know, we're gonna be getting GPT 5.6 sold any minute here.
00:04:00.719 --> 00:04:14.319
But I think the new GPT live voice model may actually be one of the biggest steps forward uh in day-to-day AI use that we've had in a very long time.
00:04:14.479 --> 00:04:16.720
That's I think for a couple of reasons.
00:04:16.800 --> 00:04:20.560
Uh, one, the models before were just really bad, right?
00:04:20.800 --> 00:04:24.160
Chat GPT's original voice mode, when it came out, it's like, oh, kind of cool.
00:04:24.240 --> 00:04:28.639
And then you had your advanced voice mode for the last like two years, and it's just really not good, right?
00:04:28.720 --> 00:04:30.959
Gemini Live, not really good.
00:04:31.120 --> 00:04:36.000
Uh, you know, Claude has a voice mode, not really good, at least, especially from the big players.
00:04:36.079 --> 00:04:47.519
There just hasn't been a great uh voice, you know, AI voice assistant that you can talk to that can pull real-time information from the web that can work with the conversation and the context that you give it.
00:04:47.680 --> 00:04:50.240
We just haven't had that, which is crazy, right?
00:04:50.319 --> 00:04:54.000
Because people are getting more and more comfortable dictating versus typing.
00:04:54.160 --> 00:04:57.360
Uh, so now this one exciting GPT Live.
00:04:57.519 --> 00:04:58.480
It is out now.
00:04:58.639 --> 00:04:59.199
It's new.
00:04:59.360 --> 00:05:00.000
Let's talk.
00:05:00.160 --> 00:05:12.399
So here's what's new OpenAI killed the kind of turn-based, you know, kind of walkie-talkie voice mode and replaced it with the new GPT Live, which is a duplex model, what they're calling it.
00:05:12.560 --> 00:05:14.800
It listens and it talks at the same time.
00:05:14.959 --> 00:05:17.519
So it also does this like back channeling like a human.
00:05:17.600 --> 00:05:21.920
So as you're talking, you know, it'll say things like, mm-hmm, got it, right?
00:05:22.000 --> 00:05:33.600
Uh, and then it redecides several times per second whether to speak, keep listening, pause, interrupt, or call a tool, like to, you know, look in the context of your uh conversation, to look on the web, etc.
00:05:34.399 --> 00:05:41.120
Also, it pulls in GPT 5.5 at the same time for web search and heavier reasoning.
00:05:41.279 --> 00:05:55.279
So, before, obviously, the voice models, you know, what they worked on, I think over the last like, you know, 2023 to 2025 was just making the voices sound more realistic and a little less latency, which is like that's cool and all.
00:05:55.439 --> 00:06:02.079
But if they don't know left from right, if they don't know what day it is, if they can't look up real-time information, what good is it?
00:06:02.240 --> 00:06:05.759
So that's what's different with this new model from OpenAI.
00:06:05.920 --> 00:06:09.199
Uh, so here is from OpenAI's blog post what they said.
00:06:09.360 --> 00:06:17.920
They said, We're launching GPT Live, a new generation of voice models that make talking with AI feel much more like having a real conversation.
00:06:18.240 --> 00:06:23.759
GPT Live is built on a full duplex architecture, meaning it can listen and speak at the same time.
00:06:23.920 --> 00:06:33.600
During conversations, GPT can show it's paying attention with phrases like, yeah, uh, engage in a quick back and forth, or just stay quiet when you need a moment to think.
00:06:33.839 --> 00:06:37.600
The result is a voice experience that is refreshingly easy to talk to.
00:06:37.759 --> 00:06:41.120
Uh, GPT Live is also our smartest voice model yet.
00:06:41.279 --> 00:06:52.160
For questions that require web search, deeper reasoning, or more complex work, it delegates to our latest frontier model behind the scenes and brings the result back into the conversation when it's ready.
00:06:52.240 --> 00:06:57.680
While it works, GPT Live can keep talking with you and maintain the flow of the conversation.
00:06:57.759 --> 00:07:08.720
Uh, at launch, GPT Live will uh use GPT 5.5 in the background as we release new frontier models, will continuously update the models used by GPT Live.
00:07:09.360 --> 00:07:12.079
So, like I said, huge news.
00:07:12.240 --> 00:07:15.040
My gosh, I've been waiting for this because I've tried.
00:07:15.279 --> 00:07:17.519
I've tried to find actual business, right?
00:07:17.600 --> 00:07:28.800
Whether I'm taking a walk or you know out in the car, or if I just like want to take a break from looking at a screen, I I've tried all every single you know voice model from all the big players.
00:07:28.879 --> 00:07:31.759
I think when Google's Gemini Live came out, it was okay.
00:07:31.920 --> 00:07:38.079
But Gemini Live actually does a bad job, strangely enough, at querying information from Google.
00:07:38.240 --> 00:07:38.800
Uh, right.
00:07:38.959 --> 00:07:39.920
So I am excited.
00:07:40.000 --> 00:07:43.920
I've gotten to play with this one a little bit, but it is brand spanking new.
00:07:44.000 --> 00:07:45.040
So who has access?
00:07:45.199 --> 00:07:46.079
Well, everyone.
00:07:46.240 --> 00:07:51.040
So if you are on the free plan, even on Chat GPT, you do have access to this.
00:07:51.199 --> 00:07:57.360
It does use a uh technically the mini version of GPT Live One, but even free users.
00:07:57.519 --> 00:08:00.800
So a billion people, yeah, a billion.
00:08:00.959 --> 00:08:04.639
That is how many weekly active users OpenAI has inside of Chat GPT.
00:08:04.720 --> 00:08:12.000
So a billion people just got a brand new, huge upgrade in what you can do with AI, right?
00:08:12.079 --> 00:08:15.839
And then on the paid tiers, you do get the full GPT Live One model.
00:08:15.920 --> 00:08:17.199
So it is live now.
00:08:17.360 --> 00:08:23.040
Make sure your app is updated, and then you just go and tap the voice button and you're off to the races.
00:08:23.199 --> 00:08:24.319
So, why is this useful?
00:08:24.480 --> 00:08:30.319
Well, it removes the dead air that made the old voice mode feel robotic, right?
00:08:30.480 --> 00:08:37.200
Uh, it you would always kind of you know interrupt each other, you're waiting, it cuts in when you don't want it to, right?
00:08:37.279 --> 00:08:40.159
It was just all voice models were really, really bad.
00:08:40.320 --> 00:08:41.840
Now, not the case anymore.
00:08:41.919 --> 00:08:49.200
And this is genuinely usable, hands-free for brainstorming, tutoring, and live QA where interruption actually matters.
00:08:49.360 --> 00:08:50.639
So, who's gonna find it valuable?
00:08:50.799 --> 00:08:56.080
Well, anyone using ChatGPT while you're driving, walking, cooking, uh, you know, packing for a trip.
00:08:56.240 --> 00:09:04.879
I think there's a lot of you know, cool use cases uh that with it being able to reason and look things up in real time that we just didn't have before.
00:09:05.120 --> 00:09:06.320
Um, right.
00:09:06.480 --> 00:09:17.519
And we've seen rumors over the last year or so that you know open AI was working on some hardware devices, those things, you know, allegedly got tabled, the Johnny Ives, you know, uh I.O.
00:09:17.759 --> 00:09:18.320
acquisition.
00:09:18.399 --> 00:09:24.240
So maybe this is the model that will power future uh OpenAI hardware when and if that comes.
00:09:24.480 --> 00:09:26.559
If so, that makes a lot of sense.
00:09:26.639 --> 00:09:33.759
Uh, because I think right now Google Home uh with Gemini is the only one that has a true smart AI home speaker.
00:09:33.840 --> 00:09:41.840
Uh, and even the difference now uh between GBT Live and Gemini Home, I mean GBT Live is in another field, at least for now.
00:09:41.919 --> 00:09:43.360
We'll see what the rest of the labs do.
00:09:43.440 --> 00:09:44.960
So uh pretty big news there.
00:09:45.120 --> 00:09:48.960
Let's go on to our second update, which is technically related.
00:09:49.039 --> 00:09:53.200
Uh, so this is kind of the developer version of this model.
00:09:53.519 --> 00:10:03.039
So from OpenAI, they also released uh the day before, so uh they released this on Tuesday two new API voice models.
00:10:03.200 --> 00:10:09.679
These are called GPT Real Time 2.1, and then the budget-friendly GPT Real Time 2.1 Mini.
00:10:10.080 --> 00:10:18.879
So this allows you to add adjustable reasoning, depth, native tool calling, and better handling of alpha numerics, um, right?
00:10:19.039 --> 00:10:24.159
Things like you know, ordering numbers, codes, etc., silence, noise, and interruptions.
00:10:24.240 --> 00:10:29.120
Uh, so open AI did say they also cut latency in this voice model by 25%.
00:10:29.679 --> 00:10:32.240
Um, and this is for developers, right?
00:10:32.320 --> 00:10:41.519
So if you uh are using the back end of the open AI platform, uh this was launched, like I said, uh this week and it is generally available.
00:10:41.600 --> 00:10:46.559
You can use it in the um open AI playground, um, but this isn't like a consumer feature.
00:10:46.639 --> 00:10:49.279
So this isn't what you're gonna, you know, log into Chat GPT.
00:10:49.360 --> 00:10:53.039
In that aspect, you're using the new GPT live model.
00:10:53.200 --> 00:10:58.879
Uh for um builders, you are using the GPT real-time now 2.1.
00:10:59.039 --> 00:11:03.279
Uh, so uh who's gonna, you know, why is this useful?
00:11:03.360 --> 00:11:04.399
Who's gonna find it valuable?
00:11:04.559 --> 00:11:13.440
I think developers, product teams who are building voice agents, uh, right, call center automation, uh, anyone working on a cost-sensitive voice startup.
00:11:13.600 --> 00:11:23.440
So even if you're not a true developer, right, this is something I'm gonna be using uh because I've wanted to build something, it probably doesn't take long.
00:11:23.600 --> 00:11:25.360
Maybe I'll do it as a live show once.
00:11:25.519 --> 00:11:31.120
It's just hard to sometimes use my phone and audio when I'm trying to do things that require audio.
00:11:31.279 --> 00:11:39.279
Anyways, um, you know, I've wanted to build a simple uh computer control agent that I can talk to and say, hey, open this, open that, right?
00:11:39.360 --> 00:11:52.240
So kind of like what I do in Codex, I dictate to Codex, but I thought, hey, it'd be cool if I just had an assistant that lives and it's always on and always listening, and I can just talk to it and it can do anything that you know the uh open AI models can do.
00:11:52.320 --> 00:11:57.600
So that's an example of something that you could build, and you can do it fairly easily just you know using codex.
00:11:57.840 --> 00:12:00.799
Uh so developers, product teams are gonna find it useful.
00:12:00.879 --> 00:12:14.159
But I mean, why it's actually useful is well, they just the lower latency and then the built-in tools mean that any voice agents that you or your team build will actually feel real time instead of laggy.
00:12:14.799 --> 00:12:16.159
All right, let's go on.
00:12:16.320 --> 00:12:18.639
Our third one, we have a new model.
00:12:18.879 --> 00:12:20.639
Is it in the frontier?
00:12:20.879 --> 00:12:22.080
Maybe we'll see.
00:12:22.240 --> 00:12:30.320
We don't have a lot of benchmarks yet, but let's talk about what we do know about Grok 4.5, the new model hot off the presses.
00:12:30.399 --> 00:12:34.879
So this is XAI, or are they called Space XAI now?
00:12:34.960 --> 00:12:35.759
I I don't even know.
00:12:35.840 --> 00:12:37.600
I think they're just XAI, right?
00:12:37.679 --> 00:12:45.360
So this is their newest flagship model, uh, tuned hard for software engineering and multi-step agentic work.
00:12:45.519 --> 00:12:54.000
Uh, so the headline claim is this XAI says that it uses four times fewer tokens than comparable models on average.
00:12:54.159 --> 00:13:00.559
So, although this isn't something that's got that's benchmarking off the charts, it is somewhat comparable.
00:13:00.720 --> 00:13:11.919
Um, you know, as an example, I'm looking at their announcement post here um on you know deep suite one one, which is I think one of the best uh software engineering benchmarks.
00:13:12.080 --> 00:13:22.720
Uh, you know, it's not great, but it's you know, at least in the tier with you know, Fable, GPT-5.5, and Opus 4.8.
00:13:22.799 --> 00:13:25.759
Uh, same thing with terminal bench, it's much more competitive.
00:13:25.840 --> 00:13:32.320
It's actually ahead of Opus 4.8, uh, and pretty close there to GPT-5.5 and Fable as well.
00:13:32.480 --> 00:13:40.000
Um, and then Sui Brand Pro, same thing, a little ahead of the um Opus models and uh GPT-5.5 there.
00:13:40.159 --> 00:13:45.840
So great uh model, specifically, I think, for software engineering.
00:13:46.000 --> 00:13:48.159
So here is what Grok says.
00:13:48.320 --> 00:13:51.039
They said today we're launching Grok 4.5.
00:13:51.200 --> 00:13:51.919
Yeah, there we go.
00:13:52.080 --> 00:13:56.639
Space XAI's see, is this XAI or Space XAI?
00:13:56.720 --> 00:13:58.240
They say it both ways, anyways.
00:13:58.399 --> 00:14:05.600
Uh, this is uh Space XAI's smartest model built to excel at coding, agentic tasks, and knowledge work.
00:14:05.679 --> 00:14:09.679
It is our strongest model ever and was trained alongside Cursor.
00:14:10.000 --> 00:14:11.519
Uh, more on that here in a second.
00:14:11.600 --> 00:14:14.080
So they say real world engineering excellence.
00:14:14.240 --> 00:14:21.919
Groc 4.5 was trained on data sets spanning knowledge and coding, science engineering, and math with both intelligent and efficient reasoning.
00:14:22.159 --> 00:14:29.360
Groc 4.5 excels at real engineering tasks and exceeds comparable leading models at these tasks.
00:14:29.519 --> 00:14:32.960
Uh, so a little bit more to know.
00:14:33.279 --> 00:14:36.240
Uh, well, right now, where can you use this?
00:14:36.399 --> 00:14:42.480
So, well, you can use it in the Grok app, but the access depends on if what plan you're on.
00:14:42.639 --> 00:14:45.120
So, free users get like hardly nothing.
00:14:45.279 --> 00:14:53.279
So, uh, if you're on a paid plan, you will have some limited use of Grok 4.5, or if you're using Grok build.
00:14:53.440 --> 00:14:54.799
So, why is it useful?
00:14:54.960 --> 00:15:00.080
Uh, well, hey, we'll see how the third-party benchmarks uh look, right?
00:15:00.159 --> 00:15:02.080
I'll be looking at artificial analysis.
00:15:02.240 --> 00:15:06.240
Uh, cost to run the test is always one of the most telling benchmarks.
00:15:06.480 --> 00:15:24.480
But if this is true, if you are able, at least on the software engineering side, able to get you know, Opus, you know, five five level um output, level intelligence at you know about a fourth of the cost, that's pretty big, right?
00:15:24.639 --> 00:15:38.240
Uh, so this is probably not gonna be a model that you know your whole team jumps in on and you're using it for all kinds of things, but this could uh in the future aim to be a specialist dev software engineering model, especially with cursor, right?
00:15:38.320 --> 00:15:44.879
So SpaceX uh acquired uh cursor as part of you know XAI, SpaceX going public.
00:15:45.200 --> 00:15:49.279
And the thing that they have that's a huge advantage is they have compute, right?
00:15:49.440 --> 00:15:51.679
Them and Meta have tons of compute.
00:15:51.759 --> 00:15:57.120
They're one of the probably the only you know, quote unquote AI companies that just have compute to spare.
00:15:57.200 --> 00:15:59.840
That's why they're running it out and they've become kind of a neo-cloud.
00:16:00.159 --> 00:16:06.159
Um, but this and that combined with cursor is going to be a big deal.
00:16:06.240 --> 00:16:11.200
So cursor working with um the XAI or Space XAI team.
00:16:11.360 --> 00:16:19.840
Uh, I'm sure that Grok and um Cursor's model, Composer uh 2.5, are gonna start to intertwine and you know learn from each other.
00:16:19.919 --> 00:16:25.840
But I I would probably be more interested in Composer 3 uh than Grok 4.5.
00:16:25.919 --> 00:16:32.960
But regardless, uh, for a lot of people, if you're out there building, you do at least have a new model to take a look at.
00:16:33.200 --> 00:16:38.639
All right, we have a lot more to get to, but first, quick break for a word from our partners.
00:16:39.120 --> 00:16:44.159
Here's what most AI tools still can do: work outside their own little box.
00:16:44.320 --> 00:16:46.639
The all new Slack bot just changed that.
00:16:46.799 --> 00:16:53.679
It's your AI teammate inside Slack, and now it can read, write, and act across the other apps your team already uses.
00:16:53.840 --> 00:16:59.200
No more tool switching, no more re-explaining yourself every time you open a new tab.
00:16:59.360 --> 00:17:05.759
One ops team at engine says the summary feature alone saves them 15 to 20 minutes of use.
00:17:05.920 --> 00:17:09.279
See what Slackbot can do at Slack.com.
00:17:12.160 --> 00:17:16.960
All right, uh, let's keep it going for our Friday features on Thursday.
00:17:17.119 --> 00:17:24.400
Yeah, during during that ad read, I don't know if you've ever done this where you have an old coffee, but you didn't know it was old, and then I almost spit it out.
00:17:24.640 --> 00:17:26.960
Um, yeah, that that'll wake you up.
00:17:27.200 --> 00:17:33.359
All right, uh our next AI feature, Notion Agents, a dedicated iPhone app.
00:17:33.680 --> 00:17:38.319
So here's what's new, why it matters, and who's gonna find it valuable.
00:17:38.480 --> 00:17:46.720
So this is a brand new standalone iPhone app from Notion, which is separate from the main Notion app.
00:17:46.880 --> 00:17:52.240
And this is out now, it is released and it's built purely for running Notion's AI agents.
00:17:52.480 --> 00:18:06.079
So you can capture by text, voice note, or photo, and then answers uh agents answer from your docs and connected tools and can take actions like creating pages, drafting updates, and running team workflows.
00:18:06.160 --> 00:18:12.319
And this is obviously model flexible, so it can work with Claude, Gemini, ChatGPT, whatever your team uses.
00:18:12.480 --> 00:18:13.759
So who has access?
00:18:14.000 --> 00:18:16.799
Well, it is live on the app store.
00:18:16.880 --> 00:18:24.880
So it is iPhone first right now, and it does run on iPad and Mac, but isn't really designed for them and no Android announcement yet.
00:18:25.359 --> 00:18:28.000
Uh so no pricing, dedicated pricing.
00:18:28.079 --> 00:18:34.960
But if you do have a paid, uh, you know, your company has a paid um notion plan, you're you're in, right?
00:18:35.039 --> 00:18:44.559
So uh it does notions, agent in AI features for the most part do require a paid plan, and those go anywhere from you know about$10 to$20 per user per month.
00:18:44.720 --> 00:18:48.160
Um, so it does tie into your existing paid plan.
00:18:48.319 --> 00:18:49.359
So why is it useful?
00:18:49.519 --> 00:18:56.480
Well, it lets you fire off workspace tasks and capture ideas from your phone instead of waiting to get to a desk.
00:18:56.559 --> 00:19:00.960
Uh, and then the voice and the photo capture kind of kill the friction of logging things on mobile.
00:19:01.119 --> 00:19:02.720
So who's gonna find this valuable?
00:19:02.880 --> 00:19:10.240
Well, if you are a team that uses Notion, uh and Notion has been, I think, crushing it on the AI side, right?
00:19:10.400 --> 00:19:17.680
I think Notion and Slack are two companies that, you know, a year and a half ago, if you would have said, hey, would they be in this kind of like second tier?
00:19:17.839 --> 00:19:26.799
I would have said, probably not, but sure enough, you know, I think Notion, Slack, and some others are, you know, really creeping up in terms of capabilities and what they can do.
00:19:26.960 --> 00:19:35.920
So, you know, if your team is on Notion and you haven't started yet taking advantage uh of their agentic features, they're actually really, really good, right?
00:19:36.000 --> 00:19:48.559
And uh met with a couple of weeks ago when I was out in San Francisco, I actually uh met with the Notion team out there, and you know, they're they're working on some uh pretty capable uh offerings, so it is something to take uh to keep an eye on.
00:19:48.799 --> 00:19:53.440
Uh so uh although I I I'm not super sure, right?
00:19:53.519 --> 00:19:56.240
I'm not a uh power notion user myself.
00:19:56.400 --> 00:19:58.799
I've used it you know off and on throughout the years.
00:19:58.960 --> 00:20:01.440
I'm not certain about the separate app.
00:20:01.680 --> 00:20:09.759
Um, so I'm not sure why it's not just built into their normal app and then having a uh an agents section.
00:20:09.920 --> 00:20:14.880
I'm sure there's some you know uh technical reason why.
00:20:15.039 --> 00:20:21.519
Um, but we'll see uh if that sticks, if it gets popular, if it doesn't, etc.
00:20:21.839 --> 00:20:22.960
All right, next.
00:20:23.359 --> 00:20:30.160
This one again, this might be for a smaller audience, but it's something that I think could make a big deal.
00:20:30.319 --> 00:20:32.160
So Google Voice AI.
00:20:32.319 --> 00:20:44.400
Yeah, so I think for a lot of small business, small business owners, if you don't know Google Voice, it is a way to just get a uh phone number, a phone line for your company in like two minutes and for super cheap.
00:20:44.559 --> 00:20:49.039
And now they're rolling out a ton of new, very helpful AI features.
00:20:49.200 --> 00:20:54.319
Uh so Google Voice has added Gemini powered take notes for me, right?
00:20:54.400 --> 00:20:59.279
So you can just tap notes mid-call and it records and transcribes everything automatically.
00:20:59.359 --> 00:21:06.000
And then after the call, you get a summary with key points and action items plus the full transcript and audio.
00:21:06.160 --> 00:21:08.720
And then the notes arrive as an email link.
00:21:08.880 --> 00:21:14.400
So here is what Google uh said about this new uh feature.
00:21:15.039 --> 00:21:21.200
So uh they say Google Voice launched over 15 years ago as a call forwarding and texting service within the US.
00:21:21.359 --> 00:21:27.519
Since then, it has evolved into a trusted global communications platform for users and businesses alike.
00:21:27.680 --> 00:21:29.680
We're taking the next step in that evolution.
00:21:29.839 --> 00:21:41.200
We're bringing the power of AI note-taking directly to your phone calls and introducing new plans designed to help users scale their businesses with a professional grade phone system at an industry leading price.
00:21:41.440 --> 00:21:46.240
So, this does require a paid Google Voice standard or premiere plan.
00:21:46.400 --> 00:21:49.359
So the free starter tier does not include it.
00:21:49.519 --> 00:21:52.880
Uh, right now it's English only for users in the US.
00:21:53.039 --> 00:21:55.119
I'm sure that will change here soon.
00:21:55.359 --> 00:21:55.839
All right.
00:21:56.000 --> 00:21:58.960
So, um, well, why is it useful?
00:21:59.119 --> 00:22:05.200
So if you Which I do know a lot of small businesses that use uh Google Voice, right?
00:22:05.279 --> 00:22:16.640
So I guess the downside is well, you they it's been behind some of the other you know, VoIP uh providers uh out there that have had some AI features uh before them.
00:22:16.880 --> 00:22:26.240
So not saying this is going to be a differentiator for businesses, but especially for small businesses that are out in the field, they have physical locations.
00:22:26.400 --> 00:22:27.920
This one's a big one.
00:22:28.079 --> 00:22:37.759
So, you know, uh it is able to, you know, turn all those phone calls into searchable records and you know, ready-made follow-up tasks with zero manual note-taking.
00:22:37.839 --> 00:22:50.960
Uh, so the standalone pricing, also, you know, if you're a solo operator, if you have a side hustle, small business, I think it's uh a way to cheaply uh bring some AI, especially if you're uh a type of business that's on the phone all the time.
00:22:51.200 --> 00:22:51.839
All right.
00:22:52.480 --> 00:22:54.799
Our next AI update.
00:22:55.039 --> 00:23:02.880
Yeah, uh Chat GPT Images, their new images 2.0 model that's been absolutely dominating since it came out.
00:23:03.039 --> 00:23:16.480
No one has been able to touch the uh OpenAI GPT Images 2 model when it comes to all benchmarks, but they may have company near the top, not passing them yet, but at least now a more serious uh contender.
00:23:16.799 --> 00:23:22.960
So this is ByteDance, their new model, CDream 5.0 Pro.
00:23:23.920 --> 00:23:29.359
So this is ByteDance's new image model built to understand design, not just generate images.
00:23:29.519 --> 00:23:36.319
So that at least from some of the samples that I've seen, uh, this is seems to be kind of where they're pushing heavy in.
00:23:36.400 --> 00:23:46.319
So adding that understanding or reasoning layer to image generation and not just having it be, you know, this diffusion technology that just decides what pixels go where.
00:23:46.480 --> 00:23:52.720
Some of the more impressive examples that I'll kind of scroll to for our live stream audience here, you know.
00:23:52.880 --> 00:24:01.119
So as an example showing like the six major types of tea, uh, right, but it has to be able to reason and think uh and break that down.
00:24:01.200 --> 00:24:06.960
There's a brewing temperature breakdown, you know, kind of a uh oxidation level.
00:24:07.119 --> 00:24:12.640
So there's a lot of reasoning and logic and thinking and planning that goes into some of these images, right?
00:24:12.799 --> 00:24:21.200
So if you think about things that might be used as an infographic for your company, you know, just strong imagery that can really support a more visual story.
00:24:21.279 --> 00:24:31.119
Uh, this, you know, is probably now in that top three, I think, you know, along with Nano Banana uh Pro and GPT Images 2.
00:24:31.279 --> 00:24:37.920
You now have Seed Dance, uh sorry, Seed Dream, not Seed Dance, Seed Dance is the video, Seed Dream is the photo version.
00:24:38.079 --> 00:24:38.480
All right.
00:24:38.559 --> 00:24:51.599
So it also adds that data and infographic visualization from dense text, interactive precision editing, uh, and stronger photorealism and uh native generation in 10 languages.
00:24:51.920 --> 00:24:57.200
But uh one of the standout features, I think, is a layer separation piece.
00:24:57.359 --> 00:25:02.480
Uh, so it edges toward it being kind of like a design tool, not just like a prompt box.
00:25:02.640 --> 00:25:14.079
So if you've ever used, you know, Canva or Photoshop and having all these different layers and being able to adjust it via layers versus as a flat image is gonna save a lot of time and give you more granular control.
00:25:14.319 --> 00:25:20.400
So this is announced and it is available inside ByteDance's own apps right now.
00:25:20.559 --> 00:25:25.599
So these are probably not apps that you use, uh, but it's in Dreamania.
00:25:26.000 --> 00:25:30.000
Uh, and then for uh Chinese users, it's in Jim Mang.
00:25:30.240 --> 00:25:36.799
Uh, so for international users, you can download the uh Dream Mania app from ByteDance.
00:25:36.960 --> 00:25:47.839
Right now, there is no public API available yet, uh, but there are some third-party platforms uh, you know, that you know are conglomerates where you can use a bunch of the different.
00:25:48.240 --> 00:25:49.759
I think fall is one of them.
00:25:49.839 --> 00:25:52.640
Is it fall or foul FAL or is it FAI?
00:25:52.720 --> 00:25:53.119
Uh right.
00:25:53.200 --> 00:25:55.440
So there's a bunch of these third-party ones that have it.
00:25:55.519 --> 00:26:06.160
So if you really do want to uh get your hands on it, if you use one of those third-party um, you know, AI image generators that have model selection, you're probably gonna see it in there.
00:26:06.720 --> 00:26:08.880
So uh why is it useful?
00:26:09.039 --> 00:26:14.319
Well, like I said, it is instantly a top three, maybe top two image model.
00:26:14.480 --> 00:26:19.279
So it turns data and dense text into clean on-brand layouts.
00:26:19.440 --> 00:26:36.400
So if for whatever reason, if you can't use Chat GPT or if you are a heavy into marketing, uh design, advertising, and you're not getting the results that you like out of Nano Banana or GPT Images 2, you now have another player to at least try.
00:26:36.559 --> 00:26:49.039
Uh so also uh multilingual teams, if you need text accurate output across different languages, designers, marketers, content teams who are producing you know infographics and branded visuals.
00:26:49.119 --> 00:26:54.400
Uh, this is definitely a new AI feature you can use now that you're gonna want to check out.
00:26:54.559 --> 00:27:01.359
All right, and then last but not least, Meta is back, back again.
00:27:01.680 --> 00:27:02.640
Meta's back.
00:27:02.799 --> 00:27:03.839
Tell some friends.
00:27:03.920 --> 00:27:04.480
All right.
00:27:04.640 --> 00:27:08.000
Uh, so let me just get this out there.
00:27:08.720 --> 00:27:14.400
Meta, when they came out with their, you know, they have their MSL, their meta superintelligence labs.
00:27:14.480 --> 00:27:19.440
They spent, you know, billions of dollars hiring all this talent, billions of dollars on compute.
00:27:19.599 --> 00:27:22.319
And then, you know, everyone thought, oh, we're gonna get a new llama.
00:27:22.400 --> 00:27:27.440
Instead, they went a different direction with the Muse, uh, that is their large language model.
00:27:27.599 --> 00:27:30.960
And there's some rumors that they're coming out with a bigger version of it.
00:27:31.039 --> 00:27:36.319
So we have Muse Spark, which is presumably uh that came out about two months ago, and that's their smaller model.
00:27:36.480 --> 00:27:43.920
And actually, at the time it came out, through a lot of benchmarks, it was maybe like a top four or five model, not anymore, right?
00:27:44.000 --> 00:27:49.920
Now it's probably in the you know eight, nine, 10 range after a lot of new updates from all the other big players.
00:27:50.160 --> 00:27:57.359
But you can't knock Meta for going from like zero to top five with their first large language model in Muse.
00:27:57.599 --> 00:28:10.160
So they've actually done the same thing with Muse image and muse video because if you're looking at Arena, uh, right, which is kind of the blind taste test for image and video models, these are top three models, right?
00:28:10.240 --> 00:28:15.440
So going from zero, never having at least under the Muse family.
00:28:15.519 --> 00:28:24.319
I know they did uh have you know their image models previously, but you know, they scrapped everything, started, you know, gotten rid of the the llama baseline.
00:28:24.559 --> 00:28:28.720
So pretty, pretty interesting results for our you know live stream audience.
00:28:28.880 --> 00:28:30.079
You can see some of these examples.
00:28:30.160 --> 00:28:31.680
We shared them in our newsletter.
00:28:31.920 --> 00:28:34.960
Pretty good, both on the photo and the video side.
00:28:35.119 --> 00:28:37.519
So here's a little bit what it is, what's new.
00:28:37.599 --> 00:28:52.480
So Muse Image is Meta Superintelligences Labs first or MSL, their first in-house image model built as an agent that can use search and code execution to self-refine its output.
00:28:52.799 --> 00:29:05.039
So, kind of the uh maybe controversial but signature feature is you can generate or manipulate images of a public Instagram account just by tagging it, right?
00:29:05.119 --> 00:29:13.920
So you kind of had this originally in Sora from OpenAI, which is no longer RAP Sora, they had this thing called cameos, right?
00:29:14.000 --> 00:29:16.319
But people had to open up, right?
00:29:16.400 --> 00:29:26.720
Like, yes, you can use my cameo, but it looks like if you're a certain level of public Instagram account, uh people can do it just by tagging you.
00:29:26.880 --> 00:29:32.960
All right, so Muse Video is built on the same foundation with native audio, but is still in development.
00:29:33.039 --> 00:29:34.799
So no release on that.
00:29:34.960 --> 00:29:49.279
Uh, so this is uh on the images side, it is free for everyday creation if you are using the meta AI app uh or meta.ai inside Instagram, WhatsApp, Facebook, Messenger, all those things.
00:29:49.440 --> 00:29:56.480
Uh, you should see uh the new um meta muse image model available.
00:29:56.559 --> 00:29:59.119
Uh, but heavy users are going to pay.
00:29:59.200 --> 00:30:06.240
So there is a kind of a free daily usage threshold, uh, and then it rolls into the new meta paid subscription plans.
00:30:06.319 --> 00:30:09.759
Uh, they didn't you know disclose the exact limit or price.
00:30:09.920 --> 00:30:13.680
Again, this is pretty new, uh, but there's no public API right now.
00:30:13.839 --> 00:30:23.200
So uh if if you're not on a paid meta plan, which I don't know anyone that's on a paid meta subscription, I know they just rolled that out a couple of months ago.
00:30:23.279 --> 00:30:32.799
Uh, you know, unless your business like runs on meta in Instagram, I don't know anyone that's actually subscribing to Meta for AI features, right?
00:30:32.880 --> 00:30:44.160
So it will be uh worth tracking when and if they come out on the API side, because at least the samples of the video model and the photo model look really good, right?
00:30:44.240 --> 00:30:51.359
But if it's fairly limited to what you can use for free, I don't know how much traction this is gonna get, right?
00:30:51.519 --> 00:31:02.079
Especially when presumably we may be getting within you know three to six months, we may be getting brand new updates to GBT images and to nano banana.
00:31:02.240 --> 00:31:10.720
So at that point, and you know, we just talked about uh the new uh you know image model uh seed dream from Bite Dance.
00:31:10.799 --> 00:31:13.680
So I don't know if I was meta, I would have came out a little more hard.
00:31:13.759 --> 00:31:18.079
Like, hey, use this thing, use it a ton because the results are pretty good.
00:31:18.160 --> 00:31:26.240
And like I said, Arena, I believe once they were released, they were both top three, top four, you know, image and video models, so fairly impressive.
00:31:26.319 --> 00:31:27.279
So, why is it useful?
00:31:27.440 --> 00:31:35.119
Well, number one, it's free right now, limited use, and it lives inside apps that billions of people already use.
00:31:35.279 --> 00:31:40.160
So, this is no new tool to learn, it's just in the apps that people use.
00:31:40.480 --> 00:31:49.039
Also, interesting to see how this plays out, but the agentic self-refinement aims for just kind of more accurate, less hallucinated images.
00:31:49.200 --> 00:31:58.960
So I think you know, the people or the groups that are gonna find this most valuable, especially early on, is if your job right now requires, right?
00:31:59.039 --> 00:32:08.559
If you're a social media manager, social media marketer, then you obviously are living inside and you know, living and breathing on your, you know, meta metrics.
00:32:08.720 --> 00:32:10.400
That's this is gonna be for you, right?
00:32:10.480 --> 00:32:19.599
So, versus having to work with an external tool, meta's obviously made this, you know, um uh this earlier technology available in their ads.
00:32:19.680 --> 00:32:21.599
I would assume that we'll see that here soon as well.
00:32:21.839 --> 00:32:35.759
So if you are a content creator, social media marketing, social media advertising, and you lay heavily uh on any of the meta, you know, Instagram, Facebook, WhatsApp platforms, this is gonna be one worth checking out.
00:32:35.920 --> 00:32:38.480
All right, so that is a wrap, y'all.
00:32:38.640 --> 00:32:40.799
Our seven big features.
00:32:40.960 --> 00:32:42.799
Let's do them in reverse order.
00:32:42.960 --> 00:33:22.880
So the new meta muse image and video models, uh, bite dances, new C Dream 5.0 Pro on the image side, Google Voice AI, great for small businesses, Notion Agents dedicated iPhone app for uh teams using uh Notion Grock 4.5 new model could be good for software engineering, open AI real time 2.1, the update to GPT Real Time 2.0, can do a lot of things agentically, and then last but not least, which I think may ultimately end up making the biggest difference for millions of business users, is the new GPT live model.
00:33:23.279 --> 00:33:26.400
So I'll probably end up doing uh a review on that.
00:33:26.559 --> 00:33:28.000
But thank you for tuning in.
00:33:28.160 --> 00:33:30.319
Thanks for the curveball again, right?
00:33:30.480 --> 00:33:34.319
Last week we had the uh July 4th observed holiday.
00:33:34.480 --> 00:33:41.359
This week we're switching it up so tomorrow we can do the show on Soul GPT 56 Soul.
00:33:41.599 --> 00:33:43.039
So make sure to join us for that.
00:33:43.200 --> 00:33:44.160
Thank you for tuning in.
00:33:44.319 --> 00:33:47.599
Hope to see you back tomorrow and every day for more everyday AI.
00:33:47.680 --> 00:33:48.640
Thanks, y'all.
