WEBVTT
00:00:01.600 --> 00:00:09.359
It seemed like this week we were gonna have a quieter week in terms of big AI releases that you can actually use today.
00:00:09.519 --> 00:00:21.600
But then late Thursday, we got a new, soon-to-be open source model that completely reset the proprietary versus open framework.
00:00:21.920 --> 00:00:29.440
So, yeah, as always, some surprises, some new updates, and some quality of life improvements.
00:00:29.679 --> 00:00:44.719
So if you spend most of your week inside of AI tools, the problem is under the hood and sometimes only on a random tweet, these companies release some big updates that change how all of these products work.
00:00:44.880 --> 00:00:47.119
And unless you spend hours, you're gonna miss those.
00:00:47.200 --> 00:00:50.640
But that's what we do on our Friday features show.
00:00:50.880 --> 00:00:58.719
So on today's show, you're gonna learn what open source model is again resetting the open versus proprietary landscape.
00:00:58.799 --> 00:01:02.799
You're gonna learn how to claude on desktop is getting a lot more useful.
00:01:02.960 --> 00:01:11.920
And you're gonna see why OpenAI made another big change to its new and very popular Chat GPT Work Desktop app.
00:01:12.319 --> 00:01:13.680
All right, let's get into it.
00:01:13.840 --> 00:01:14.959
Welcome to Everyday AI.
00:01:15.040 --> 00:01:17.040
My name's Jordan Wilson, and we do this every day.
00:01:17.200 --> 00:01:25.280
This is your daily live stream podcast and free daily newsletter helping business leaders like you and me keep up with the nonstop avalanche of AI updates.
00:01:25.439 --> 00:01:27.280
I tell you what matters, what doesn't.
00:01:27.439 --> 00:01:30.640
You use that information to be the smartest person in AI at your company.
00:01:30.799 --> 00:01:32.319
Everyone's like, wow, how'd you do this?
00:01:32.480 --> 00:01:38.319
And you're like, well, I went to everyday your everydayai.com and signed up for the free daily newsletter.
00:01:38.400 --> 00:01:41.840
And well, you listen on the podcast and uh maybe on the live stream.
00:01:41.920 --> 00:01:42.959
So thanks for doing that.
00:01:43.200 --> 00:01:48.319
And if you are looking for that AI news, make sure to check it out in today's newsletter.
00:01:48.640 --> 00:02:04.799
All right, let's get straight into it and talk about our first new update, which I'm like, bless up, it's about time because I don't know about you, but with all these desktop apps, I've been trying to use them just like never leaving.
00:02:04.879 --> 00:02:10.719
So when I go into, as an example, codec slash chat GPT work, I try to not leave, right?
00:02:10.879 --> 00:02:12.639
Really change how I do my work.
00:02:12.800 --> 00:02:18.800
And one of the big downsides with Claude Desktop is you really couldn't do that until this week.
00:02:18.960 --> 00:02:24.080
That's because Anthropic did finally add their new desktop browser.
00:02:24.159 --> 00:02:27.439
So yeah, bringing a browser inside of Claude Desktop.
00:02:27.520 --> 00:02:40.479
So here's what's new Anthropic added a built-in browser to the Claude Code Desktop app, which lets Claude open external websites and actually click, type, and interact inside of them while it works on your code.
00:02:40.719 --> 00:02:50.159
So Claude can now open whatever you need inside of the Claude app, whether that's your API docs, a bug tracker, dashboard, whatever it may be.
00:02:50.319 --> 00:02:53.919
So it's not just reading the pages inside of there, it can act on them.
00:02:54.080 --> 00:03:01.919
Uh, Anthropics says safety is layered in as classifiers review every click and keystroke Claude makes on external sites.
00:03:02.159 --> 00:03:07.360
And the first time Claude touches a new site, you get a permissions card to allow or deny it.
00:03:07.439 --> 00:03:09.680
Uh, every site needs its own approval.
00:03:09.840 --> 00:03:11.520
So that's one of the things for me.
00:03:11.599 --> 00:03:13.360
I'm like, uh, that's a downside.
00:03:13.439 --> 00:03:26.639
Uh, but overall, this is a big step forward for anthropic, putting it more in parody, uh, mainly with codec slash chat GPT work and also cursor, which I've always said were the top two in terms of super app.
00:03:26.960 --> 00:03:32.159
Uh so now uh enthropic claw desktop can finally enter the conversation and we'll see.
00:03:32.240 --> 00:03:42.800
Presumably, uh, we'll have Windows uh with their new uh super app Microsoft Copilot entering, we think it should be probably by next month.
00:03:43.120 --> 00:03:59.199
So uh who has access, anyone running the Claude Code or Matt, uh, you know, I just call it the Claude Desktop app because it's not just Claude Code, uh, but the Claude Desktop app on Mac in Windows, and you do need a paid plan to take advantage of these features.
00:03:59.439 --> 00:04:05.280
Um, so an important choice though that they made the profile uses a completely clean profile.
00:04:05.439 --> 00:04:09.199
So none of your saved logins, none of your history, uh, etc.
00:04:09.439 --> 00:04:22.800
And that's one of the big updates that actually OpenAI rolled out to their Chat GPT work uh last week as they unveiled the non-technical version of uh codecs called the ChatGPT work.
00:04:23.040 --> 00:04:26.240
Well, it can import all of your Chrome history, everything.
00:04:26.480 --> 00:04:31.759
So a much more robust browser on the codex slash chat GPT work side.
00:04:31.920 --> 00:04:42.240
However, at least even for me personally, the way that I've changed my work, this was one of the big things that I just like I wasn't really touching Claude Desktop anymore.
00:04:42.480 --> 00:04:44.399
Uh, at least not very much, right?
00:04:44.480 --> 00:04:49.920
I still use it for you know adversarial, uh kind of working with codecs, right?
00:04:50.000 --> 00:04:57.439
I do some more advanced things that I'll I'll use claude desktop for, like I, you know, running GPT-5.6 soul inside of Claude Desktop.
00:04:57.600 --> 00:05:09.120
Conversely, I'm running Fable inside of Codec slash chat GPT work, but I really wasn't using Claude Desktop as much as I was, maybe like three or four months ago, mainly because of all the updates that Codex has.
00:05:09.279 --> 00:05:13.040
But at least for me, this one brings it a little bit more back into the conversation.
00:05:13.199 --> 00:05:14.560
So who's gonna find this valuable?
00:05:14.639 --> 00:05:19.759
I think Deb's obviously already working inside Cloud Code, uh, security teams, uh, right.
00:05:19.920 --> 00:05:31.120
Or if you're a general knowledge worker and you've been doing some things inside of Cloud Code and you're like, wow, it would be nice to bring, you know, some additional work that I do on the web into the fold.
00:05:31.279 --> 00:05:33.839
Uh, I think that's gonna be a big one for you.
00:05:34.319 --> 00:05:37.600
All right, let's go to our next update.
00:05:37.759 --> 00:05:45.759
So this one is more of a quality of life one that I'm like, yes, uh, I need this, and I think most people will enjoy it.
00:05:46.079 --> 00:06:00.000
So OpenAI launched universal search in Chat GPT, which lets you search in one place for all of your past chats, projects, images, and documents, all from one search box in the sidebar.
00:06:00.079 --> 00:06:06.480
Uh, so you can filter by content type uh and clicking a request that opens the chat project or file directly.
00:06:06.639 --> 00:06:10.879
So, yeah, this just launched a couple of days ago, kind of an under the hood.
00:06:10.959 --> 00:06:21.120
But again, this is one that should hopefully bring a big quality of life for the uh you know hundreds of millions of weekly active chat GPT users.
00:06:21.360 --> 00:06:27.279
The good thing this rolls out to everyone, whether you have a free pan, a free plan, uh paid plan, etc.
00:06:27.439 --> 00:06:32.000
And it did already roll out uh on the web and iOS and Android.
00:06:32.079 --> 00:06:33.040
So that's the other good thing.
00:06:33.199 --> 00:06:35.199
This also works inside of ChatGPT.
00:06:35.439 --> 00:06:45.360
So if you're anything like me, uh you spend a lot of time, regardless of what platform you're using, trying to find old chats, especially on the web.
00:06:45.519 --> 00:06:52.720
The good thing is like ChatGPT work and codecs are great at this because one thread can read every single other thread and direct you.
00:06:52.879 --> 00:07:02.639
But at least on the web, this has always been something I've struggled with because sometimes chats, uh, you know, these searches will only search normal chats, or they might not search through your project.
00:07:02.800 --> 00:07:08.160
So it is good to have a single universal chat, just makes Chat GPT much more accessible.
00:07:08.399 --> 00:07:10.240
So why is it useful?
00:07:10.399 --> 00:07:14.800
Well, I mean, your Chat GPT history just stops becoming a graveyard, right?
00:07:14.879 --> 00:07:22.399
That analysis that you ran in March that you put a ton of time in, and maybe you use some of it, but there was a lot of it that you were like, hey, we could benefit from this.
00:07:22.480 --> 00:07:23.199
Where is it?
00:07:23.360 --> 00:07:28.160
Uh, you know, the image that you did last week, the the contract you uploaded in May, right?
00:07:28.319 --> 00:07:33.199
It's all findable in seconds versus having to scroll forever, right?
00:07:33.360 --> 00:07:35.759
I would have loved this like a year or so ago.
00:07:35.920 --> 00:07:51.519
There were actually a few instances, uh, both in Chat GPT and Claude, where I had a very important, uh, couple of very important chats that I worked on that I literally couldn't find because sometimes that uh, you know, searching will literally only search the thread name.
00:07:51.680 --> 00:08:05.759
So in multiple instances, I've had to go through, export all of my chats, import them in, use a thinking mode just to find uh, you know, a certain chat that was at least for me very valuable or very useful that I needed to repurpose or reuse.
00:08:05.920 --> 00:08:06.959
So uh this is good.
00:08:07.040 --> 00:08:09.519
And like I said, this is rolling out to anyone.
00:08:09.680 --> 00:08:11.279
And who's gonna find this useful?
00:08:11.360 --> 00:08:17.680
I mean, if you're a power user of Chat GPT, this is an instant just quality of life upgrade.
00:08:18.000 --> 00:08:25.439
All right, our next one, not everyone uses superhuman, but um, I was actually surprised.
00:08:25.519 --> 00:08:33.279
So, to do, you know, to help me plan for this show, I also see, you know, hey, in our newsletter, what are those stories that people are clicking on the most?
00:08:33.360 --> 00:08:34.480
And I was actually surprised.
00:08:34.559 --> 00:08:37.759
This is one of our more clicked-on stories this past week.
00:08:37.919 --> 00:08:42.879
So apparently a lot of you all use superhuman for your mail.
00:08:42.960 --> 00:08:47.679
So, well, they launched a helpful new feature if you are a superhuman male user.
00:08:47.759 --> 00:09:04.639
And if you don't know what superhuman is, well, it's kind of confusing now because it used to just be the uh, you know, the I guess minimal and fast email program, but now superhuman itself, uh, the name is you know okay, let me just explain this a little better.
00:09:04.879 --> 00:09:21.360
So Grammarly, you know, a popular uh writing tool that a lot of people used, technically acquired superhuman, but they applied the superhuman uh name to all of Grammarly's other tools.
00:09:21.519 --> 00:09:29.039
So it is a little confusing now, but uh here's what's new if you do use superhuman with their new auto drafts.
00:09:29.120 --> 00:09:37.120
So superhuman launched a new version of auto drafts, which writes replies to your emails in your voice before you even open your inbox.
00:09:37.360 --> 00:09:47.360
So it does draft automatically two things responses to messages waiting on you, and follow-ups for emails other people haven't answered yet.
00:09:47.519 --> 00:09:52.159
So you open your inbox and the replies are sitting there for your review and to hit send.
00:09:52.320 --> 00:09:55.200
So the actual new thing here is the engine running it.
00:09:55.279 --> 00:10:04.240
So it now runs on frontier models from Anthropic and OpenAI, replacing an old GPT 3.5 version.
00:10:04.399 --> 00:10:10.799
Yeah, apparently they were still using GPT 3.5, which is maybe why the results weren't that good.
00:10:10.960 --> 00:10:15.279
So that's why now uh the drafts can actually sound like you.
00:10:15.440 --> 00:10:19.600
So right now, this is for paid plans only on business and enterprise.
00:10:19.759 --> 00:10:27.279
So you can set it up on the desktop and it syncs to mobile, and it also drafts the drafts sync with Gmail and Outlook.
00:10:27.519 --> 00:10:29.360
Uh, so who's gonna find it useful?
00:10:29.519 --> 00:10:36.399
Well, obviously, if you're a paid superhuman user struggling to keep up with emails, this is a good quality of life upgrade.
00:10:36.559 --> 00:10:40.399
And you know, at first I was almost like, should I even include this?
00:10:40.480 --> 00:10:50.799
Because this is one of those things that you can very easily do inside of uh Claude, you can very easily do inside of Chat GPT Codex Chat GPT work.
00:10:51.039 --> 00:10:52.240
But I'm like, yeah, there.
00:10:52.480 --> 00:10:55.840
I mean, if you're using these tools, keep using them, right?
00:10:56.000 --> 00:11:05.679
Uh, but for me, it it is technically very easy to build this inside of like ChatGPT work or codex, and it can send them all there.
00:11:05.759 --> 00:11:10.399
So part of me was like, all of this already exists and probably platforms you're already using.
00:11:10.480 --> 00:11:15.360
But if you all already a power superhuman user, uh, this is for you.
00:11:15.519 --> 00:11:23.039
Speaking of app specific uh updates, this one I'll probably use because I'm a heavy Spotify user.
00:11:23.200 --> 00:11:26.639
And if you are too, this is one that you might enjoy.
00:11:26.879 --> 00:11:39.759
So Spotify launched Talk to Spotify, which lets you have an actual conversation with the app by voice or text to control your music, discover new stuff, and also ask about your own listening history.
00:11:39.919 --> 00:11:44.480
So the cool thing here is well, it's back and forth, not just one-shot commands.
00:11:44.639 --> 00:11:55.360
So you can ask for an artist, then keep steering, and you can say just hit his recent stuff or make it more upbeat, and then you can save and cue and follow right on from the conversation.
00:11:55.519 --> 00:11:57.840
Uh, so who has access right now?
00:11:57.919 --> 00:12:05.759
It is in beta, all right, but it's rolling out to people in the US, Ireland, and Sweden on iOS and Android.
00:12:05.840 --> 00:12:11.519
So, yeah, you do have to be a paid uh user in one of those countries, like I am.
00:12:11.600 --> 00:12:15.360
I haven't actually checked this one yet because this one just came out, but I'll be using this one.
00:12:15.600 --> 00:12:17.440
So uh, why is it useful?
00:12:17.600 --> 00:12:19.600
Well, the differentiator is your data.
00:12:19.679 --> 00:12:23.919
So Spotify knows your playlist, your favorite artists, and your repeat listens.
00:12:24.000 --> 00:12:27.279
So it answers things like when did I first listen to this song?
00:12:27.440 --> 00:12:31.600
And you know, no general chatbot, at least right now, can do that, right?
00:12:31.759 --> 00:12:39.759
Although Spotify does have some integrations with like ChatGPT as an example, it doesn't have that level of granular data.
00:12:39.919 --> 00:12:54.480
Um, so this one for me, and the reason why I think it's very helpful, if you have a paid Spotify account and multiple people in your family use it, maybe if you have kids, you have you know people with wildly different tastes in music, but everyone's using the same Spotify account, right?
00:12:54.559 --> 00:13:03.600
Like if you have uh, you know, an Alexa or you know, you know, Home Speaker, like that's what I do, and everyone's listening to all different kinds of music on my Spotify account.
00:13:03.679 --> 00:13:10.399
So, you know, Spotify used to have, you know, and they still do, it's it's these weekly playlists that update based on your history.
00:13:10.639 --> 00:13:14.240
So I used to love these playlists, they would update every Friday.
00:13:14.320 --> 00:13:24.799
I would literally used to open it every Friday and be like, oh my gosh, I can't wait for my, you know, the two playlists are called uh, I think like Discover Weekly and Release Radar based on like your uh listening history.
00:13:24.960 --> 00:13:27.759
And you know, so recently mine is just all off.
00:13:27.919 --> 00:13:35.679
So this is one I'm gonna go in there and try to kind of you know discover new music based on the genres that I actually listen to.
00:13:35.840 --> 00:13:46.000
So this one obviously maybe a little bit more on the personal side, uh, but I think it's a pretty big update, especially if you are a power Spotify user like I am.
00:13:46.480 --> 00:13:51.679
All right, our next one in this one, if you are missing Sora from OpenAI, right?
00:13:51.840 --> 00:14:07.279
Remember uh about five, five or so months ago, you know, OpenAI kind of announced that they were killing off all their side quests and really focusing, you know, on their core products, which in theory has been great, and the company is growing just exponentially.
00:14:07.440 --> 00:14:15.360
Uh, but one of the downsides is there was these fantastic products that are now just shelfware, and one of those was Sora.
00:14:15.600 --> 00:14:24.159
And one of the most popular features of Sora was being uh able to essentially import your own self, your own avatar, right?
00:14:24.240 --> 00:14:29.440
You would have to verify your face, and then you could create videos with yourself in them.
00:14:29.679 --> 00:14:34.480
Well, uh Gemini Omni has finally rolled that out.
00:14:34.639 --> 00:14:38.000
So this is technically a new feature in two different parts.
00:14:38.240 --> 00:14:47.919
So, first, Google launched personal avatars inside of Gemini, which lets you record your face and voice, then generate AI videos of yourself on demand.
00:14:48.080 --> 00:14:54.960
So you just drop your avatar into prompts by typing the plus at sign uh and then your username.
00:14:55.039 --> 00:15:05.120
So, as an example, you would type, you know, create a video of at me uh singing with an orchestra, and then your likeness becomes a reusable asset inside of Gemini.
00:15:05.519 --> 00:15:12.000
So, right now you do have to have a paid account and be in the US and 18 or older.
00:15:12.159 --> 00:15:16.960
So, yeah, all the you know, EU, UK, uh not available right now.
00:15:17.120 --> 00:15:23.120
So, uh, which is all you know, people are always like, Why doesn't you know the EU get this?
00:15:23.279 --> 00:15:30.559
Well, the European exclusion is almost always about you know regulatory caution around biometric likeness, right?
00:15:30.639 --> 00:15:40.799
That's why so many features, especially ones like this that are hyper-personalized uh for your own likeness or others, you know, don't usually roll out sometimes at all to the EU.
00:15:41.360 --> 00:15:48.240
Uh, so and and so it's not only inside of Gemini, but it's also uh inside of Google Vids.
00:15:48.559 --> 00:15:52.960
So here is what they said about using this new feature inside of Google Vids.
00:15:53.200 --> 00:16:00.799
They said today we're rolling out two new updates to Google Vids to make it easier than ever to create, edit, and personalize your videos.
00:16:01.039 --> 00:16:03.519
Gemini Omni and personal avatars.
00:16:03.679 --> 00:16:06.960
Gemini Omni takes the hard work out of the editing process.
00:16:07.120 --> 00:16:12.320
So generating and refining high-quality clips is as easy as writing a simple prompt.
00:16:12.399 --> 00:16:18.480
And with personal avatars, you have an entirely new way to star in videos without setting up a camera.
00:16:18.639 --> 00:16:37.519
So, yeah, if you missed Google's big Gemini announcement, uh Gemini Omni announcement back in May, uh essentially it is the new family of VO models, but just much more powerful where their Vio, you know, so VO3, VO3.1, we're technically just video models, right?
00:16:37.679 --> 00:16:39.840
Gemini Omni is much more than that.
00:16:40.000 --> 00:16:43.279
It produces video, but it is like a world model.
00:16:43.440 --> 00:16:49.679
And I think it's much better at editing scenes uh than any other platform out there.
00:16:49.759 --> 00:16:51.440
Uh, so this is a pretty big one.
00:16:51.519 --> 00:16:55.679
I think uh, you know, who has you know, why is it useful?
00:16:56.080 --> 00:17:04.240
Because this is like the talking, you know, so many talking head videos that just need like background B-roll, right?
00:17:04.319 --> 00:17:09.440
So if you're someone in your company that's whether it's for internal or external purposes, right?
00:17:09.599 --> 00:17:15.359
I wouldn't use these to actually do like a full avatar, like talking head video.
00:17:15.519 --> 00:17:16.319
That's not the thing.
00:17:16.400 --> 00:17:19.680
But if you or someone else, well, like me, right?
00:17:19.759 --> 00:17:22.079
I'm constantly talking head video right now.
00:17:22.160 --> 00:17:23.519
I am a talking head video.
00:17:23.599 --> 00:17:37.039
Uh, you know, I don't necessarily not part of my brand or the everyday AI brand necessarily, but when I'm yapping about all these things, I could very well instead put an avatar of me working on these things.
00:17:37.119 --> 00:17:44.640
So you don't just have to, you know, look at my uh face made for radio and just be like, all right, guy, you know, put something else on the screen.
00:17:44.720 --> 00:17:47.119
So, but I think there's a ton of great use cases for this.
00:17:47.200 --> 00:17:54.720
So if you are in LD, uh, you know, training, uh, marketing, content creation, show social media, anything like this.
00:17:54.960 --> 00:18:03.440
And if you have a CEO as an example that's hard to nail down and you are trying to get more video content out there or something like this, right?
00:18:03.599 --> 00:18:09.359
Again, if you go through the proper channels, do all the approvals, data security, privacy, all that good stuff.
00:18:09.440 --> 00:18:14.880
But now this is such a a weapon to have in your creative arsenal, right?
00:18:15.039 --> 00:18:29.200
I would have loved to have this like 18 years ago at one of my first jobs where I was creating a lot of video and usually it was just kind of talking head of the CEO, and I would try to get some of this more like B-roll stuff, and it was just sometimes hard.
00:18:29.440 --> 00:18:40.079
So I think if you are a creative, if you are trying to have a a stronger uh, you know, presence uh on social media for your company, but it's a little hard.
00:18:40.240 --> 00:18:40.720
This is great.
00:18:40.799 --> 00:18:46.160
So yeah, content creators, marketers, educators, um, you know, is is gonna be huge.
00:18:46.400 --> 00:18:54.559
But I think the simple framing of this is it's kind of like a personal version of Hey Jen inside of Gemini, right?
00:18:54.640 --> 00:19:00.480
So Avatar video just went from being kind of a specialist tool uh to now going mainstream.
00:19:00.720 --> 00:19:07.839
All right, speaking of mainstream, let's tackle our next update because this one is very mainstream.
00:19:08.079 --> 00:19:19.039
So uh the new Chat GPT Work Desktop app has a big update that I think a lot of Chat GPT power users are going to enjoy.
00:19:19.279 --> 00:19:26.559
All right, so uh I'm just gonna go ahead and read uh what Tebow, the head of product at OpenAI, posted.
00:19:26.720 --> 00:19:29.039
It's a little easier just to read it, and this is brand new.
00:19:29.119 --> 00:19:33.759
I've been playing around with it a little bit today, but yeah, it hasn't even been out a full day yet.
00:19:33.920 --> 00:19:41.519
So uh Tebo said, uh evening, we've got we've gotten lots of great feedback on the new Chat GPT desktop app.
00:19:41.680 --> 00:19:46.079
So the work app, uh, which we didn't get totally right on the first try.
00:19:46.319 --> 00:19:48.640
And as a result, we made some changes.
00:19:48.799 --> 00:19:55.119
Number one, chat GPT conversation history and projects are now visible in the sidebar.
00:19:55.279 --> 00:20:01.039
Also, your chat and work history now sync across web, mobile, and desktop.
00:20:01.119 --> 00:20:04.079
Uh, local task will stay on your computer.
00:20:04.240 --> 00:20:15.359
Uh, then you can now easily switch between chat and work modes inside chat GPT on desktop, which is now also consistent with how it shows up on web and mobile.
00:20:15.519 --> 00:20:18.480
Nothing is changing for users on codex mode.
00:20:18.559 --> 00:20:22.000
Uh, Tebow says it's still the OG and best at what it does.
00:20:22.319 --> 00:20:24.640
So, what does this mean?
00:20:24.880 --> 00:20:33.839
Well, to actually say what it means, and you know, Tebow did kind of say it there, he said, yeah, we made some, we didn't totally get it right on the first try.
00:20:34.000 --> 00:20:36.880
And that kind of explains the update.
00:20:37.119 --> 00:20:49.839
Um so long story short, if you missed this, I just did uh an episode on this on Wednesday where we went over uh Chat GPT work, what it is, all that good stuff.
00:20:49.920 --> 00:20:55.839
But chat GPT work is essentially the non-technical version of Codex, right?
00:20:56.000 --> 00:20:59.039
OpenAI's uh autonomous desktop agent.
00:20:59.200 --> 00:21:04.960
Uh, but there is a chat GPT work on the web, and there is a chat GPT work on the desktop.
00:21:05.200 --> 00:21:14.799
So the problem was it the chat GPT work didn't exactly sync up very well with what you were doing on Chat GPT on the web.
00:21:15.039 --> 00:21:19.039
It was kind of this separate pop-up that came up in the right hand corner.
00:21:19.200 --> 00:21:32.880
So your chats were kind of there, but it just wasn't really intuitive to use because you would have all of your uh essentially your tasks in your projects that lived or started with chat GPT workslash codecs on the left sidebar.
00:21:32.960 --> 00:21:39.920
And then your chat GPT history was kind of this orphan page on the right hand side that wasn't really attached to anything.
00:21:40.079 --> 00:21:48.720
And then the big thing that people were like is hey saying, like, hey, I love Chat GPT, but I run everything inside of projects, and projects did not sync.
00:21:48.880 --> 00:21:58.880
So essentially, OpenAI changed and technically fixed all of that because now not only do your projects sync to the desktop version of ChatGPT work, right?
00:21:58.960 --> 00:22:01.279
But they're also in Left hand sidebar.
00:22:01.359 --> 00:22:09.680
And the other good thing is from an aesthetic standpoint, now it does look the exact same as it does on the web, which I think is going to help people.
00:22:09.839 --> 00:22:12.160
And it looks the exact same as it did on mobile.
00:22:12.240 --> 00:22:19.200
So the first variation of this it looked very similar uh on the web and on the mobile app.
00:22:19.279 --> 00:22:24.160
But then when you open the chat GPT work app, it looked and functioned completely different.
00:22:24.319 --> 00:22:34.240
So this is a great update uh for people who are trying out uh ChatGPT work, or if you're like me, right, I've been using Codex since day one.
00:22:34.400 --> 00:22:48.240
And the good thing is, well, now you have that ChatGPT work experience, which is just a very similar version of Codex, and then you have your familiarity with all of your uh normal Chat GPT chats and your uh projects.
00:22:48.400 --> 00:22:53.519
So if you're looking at this, if if you scroll down, it's all going to be under your recents.
00:22:53.680 --> 00:23:06.000
So whether you uh start a new task inside of ChatGPT work that's not attached to a project, it will go to your recents, and that's also where all of your new chats inside of Chat GPT will live.
00:23:06.160 --> 00:23:16.640
So if you do start a new chat inside of chat GPT.com, you don't attach it to a project, and then you go to ChatGPT work on the desktop, it will be there, which is great in the same place under recent.
00:23:17.279 --> 00:23:22.559
All right, our last piece of AI news, and this is technically the biggest one.
00:23:22.640 --> 00:23:35.440
So, not just a new heavyweight on the soon to be open source, and I'll explain that, but this one is definitely resetting the open versus close model paradigm.
00:23:35.599 --> 00:23:42.160
Yeah, we all thought it was the uh GLM 5.2 that was gonna do this, not anymore.
00:23:42.319 --> 00:23:49.039
Get ready because you're gonna be hearing a lot, especially if you are an AI large language model dork like I am.
00:23:49.200 --> 00:23:58.079
A lot of the conversation, I'm guessing for the next month or two is gonna be set around Kimmy K3, and this has huge implications.
00:23:58.160 --> 00:24:01.279
So, let me first explain what it is, what's new.
00:24:01.519 --> 00:24:17.839
So, uh Chinese AI startup Moonshot released Kimi K3, a 2.8 trillion parameter model that is now the largest open model ever built with a native vision mode and 1 million token context window.
00:24:18.160 --> 00:24:31.519
So here's the thing it came in at third place overall on the artificial analysis intelligence index, not far behind Claude Fable 5 and GPT 5.6.
00:24:32.000 --> 00:24:36.880
So uh that alone is mind-boggling, right?
00:24:36.960 --> 00:24:42.319
Uh, because we've been seeing this race just go back and forth, back and forth.
00:24:42.480 --> 00:24:53.119
And we thought that when Anthropic released Mythos and Fable 5, that this was essentially a new category, a new tier that no one else was ever going to touch.
00:24:53.279 --> 00:25:12.720
So not only uh, you know, about a month later did GPT 5.6 enter that category and on many of the most important benchmarks pass Fable 5, but now we have Kimmy K3, a soon-to-be open model that has not only entered the conversation, but it is actually surpassed.
00:25:12.799 --> 00:25:20.640
Yes, an open model has surpassed well both Fable and GPT 5.6 on many important benchmarks.
00:25:21.279 --> 00:25:30.160
Uh, so uh the best open model in the world now sits at three behind the two closed flagships.
00:25:30.319 --> 00:25:42.160
And the important thing that is thrusting this all back into the conversation is well, Anthropic is supposed to be pulling Fable 5 from subscriptions this Sunday.
00:25:42.400 --> 00:25:55.920
So not only do they have this continued pressure uh from GPT 5.6, but now there is an open model that you is gonna be better, even if you're paying$200 a month like I am, you're not gonna get access to Fable 5.
00:25:56.000 --> 00:26:09.440
There are obviously rumors that maybe today or maybe uh early next week we might be seeing Opus five, and maybe that will even reset and be much closer to Fable level than currently Opus 4.8.
00:26:09.599 --> 00:26:14.640
All right, anyways, let's talk a little bit more about Kimi K3, who has access to all that stuff.
00:26:14.880 --> 00:26:22.319
One important thing to note though, right now it's not technically an open source model, but it will be soon.
00:26:22.400 --> 00:26:28.000
Uh, Moonshot did say that they will be releasing the weights, so it will be an open weights model.
00:26:28.160 --> 00:26:36.720
Uh, they just haven't released the weights yet, but it just came out literally hours ago, and everything else that Moonshot has released in the past has been open.
00:26:36.880 --> 00:26:41.039
So this is live today in the Kimi app on Kimi.com.
00:26:41.119 --> 00:26:47.119
That's K-I-M-I, uh, and also in the Kimi Work Desktop app, as well as through the API.
00:26:47.440 --> 00:26:53.920
So they did say that they will release the full model weights by July 27th.
00:26:54.000 --> 00:26:55.920
So that's within like a week and a half.
00:26:56.000 --> 00:26:59.440
So at that time, anyone can download and self-host this.
00:26:59.519 --> 00:27:03.359
Obviously, it is a large 2.8 trillion parameter model.
00:27:03.519 --> 00:27:12.319
So uh if if you think you're gonna post this on normal consumer hardware, no, unless you've spent like 20 or 30,000 on your setup.
00:27:12.480 --> 00:27:14.880
Uh, so this is more for enterprise companies.
00:27:15.039 --> 00:27:18.960
Well, if you have the uh the GPUs, you can do this.
00:27:19.200 --> 00:27:22.079
So who's gonna find this valuable?
00:27:22.240 --> 00:27:24.079
Actually, no, let's first talk about why it's useful.
00:27:24.240 --> 00:27:30.559
Well, from the benchmarks alone, what we're seeing is its specialty is in long autonomous work, uh, right?
00:27:30.640 --> 00:27:47.680
So they kind of shared two different case studies where K3 designed and verified a working chip in a single 48-hour unsupervised run, and it reproduced an Astrophysics, uh, sorry, astrophysics research result in about two hours that they say would have normally taken a researcher, one to two weeks.
00:27:47.759 --> 00:27:58.960
And then the benchmarks obviously shows that it beats many models tested, including Fable 5 on web research, uh automation, spreadsheets, document reading, and more.
00:27:59.279 --> 00:28:06.240
And the crazy thing to me is on arena, which goes head to head, uh blind taste tests.
00:28:06.319 --> 00:28:13.279
One of the things that anthropic has really kind of dominated this space is front-end design, right?
00:28:13.440 --> 00:28:20.640
So um on the design arena, GPT-5.6 uh surpassed anthropic's models, including Fable 5.
00:28:20.720 --> 00:28:28.160
And then on the LM arena, which is kind of like user preference, this model is now number one, Kimi K3.
00:28:28.559 --> 00:28:37.039
So it's interesting because that has always been one of uh, you know, anthropic's kind of niches that they've owned, like front-end design.
00:28:37.200 --> 00:28:44.160
And now on the two most important front-end design benchmarks, they are no longer number one in either.
00:28:44.319 --> 00:28:48.000
So I'm personally excited to see what anthropic is gonna cook up.
00:28:48.160 --> 00:28:58.960
Maybe we'll see this with the open five drop, but I would assume that their next model, maybe they've let their foot off the gas in terms of front-end design because they've been so far ahead of everyone for so long.
00:28:59.119 --> 00:29:11.039
So I do and would assume that Opus five is probably gonna surpass Fable Five, at least in those areas, because that's something I'm sure that Anthropic is gonna be feeling a lot of pressure on.
00:29:11.200 --> 00:29:13.440
Aside from you know, pulling Fable 5.
00:29:13.519 --> 00:29:15.039
We'll see if they extend it again.
00:29:15.119 --> 00:29:30.000
Uh, but on the subscription package between GPT 5.6 and uh now Kimi K3, uh, yeah, they're gonna have to maybe justify why people are gonna be subscribing, but a lot of people are pointing to that just means we're getting an Opus V here soon.
00:29:30.160 --> 00:29:32.880
So, who's gonna find this new Kimmy K3 valuable?
00:29:32.960 --> 00:29:39.440
Uh, enterprises and developers who want frontier adjacent agents without that frontier pricing or vendor lock-in.
00:29:39.680 --> 00:29:59.839
Uh, so I mean, the crazy thing is, I mean, this is literally now an open, soon-to-be open model from a now from a Chinese uh you know startup that now benches ahead of Opus 4.8, and it is in the same tier, although technically lower than Fable 5 and GPT-5.6 soul.
00:30:00.000 --> 00:30:05.359
All right, so an interesting one here resetting the open versus closed race.
00:30:05.519 --> 00:30:08.079
And you know what it means for all of us, y'all?
00:30:08.240 --> 00:30:18.960
We're gonna continue to get more and more models, better models, and hopefully at cheaper prices now that we have an open model, presumably pushing the cost down.
00:30:19.039 --> 00:30:23.039
So, this one here with Kimi K3, definitely more of an enterprise play.
00:30:23.119 --> 00:30:30.480
Like I said, this is not really for inner uh for consumer hardware, although the prices are also much cheaper via the API.
00:30:30.640 --> 00:30:35.200
All right, so that's a wrap, a lot new that we went over today.
00:30:35.359 --> 00:30:36.880
I hope this one was helpful.
00:30:36.960 --> 00:30:44.240
Like I said, we do this every single Friday, our Friday feature show, where we bring you the latest and the greatest of what's new that you can actually use today.
00:30:44.400 --> 00:30:51.279
I was actually kind of bummed because we also had from Google a Gemini Notebook that came out, but it wasn't available to all paid users.
00:30:51.440 --> 00:30:52.160
So that's the thing.
00:30:52.240 --> 00:30:54.799
We do this if you have a base paid account.
00:30:54.960 --> 00:30:58.640
These are the things that you can use today to grow your company and career.
00:30:58.799 --> 00:30:59.039
All right.
00:30:59.119 --> 00:31:03.920
If this was helpful, do me a favor, please subscribe on the podcast and go to your everydayai.com.
00:31:04.079 --> 00:31:05.519
Sign up for the free daily newsletter.
00:31:05.599 --> 00:31:06.319
Thanks for tuning in.
00:31:06.400 --> 00:31:09.680
We'll see you back Monday and every day after that for more everyday AI.
00:31:09.839 --> 00:31:10.720
Thanks, y'all.
00:00:01.600 --> 00:00:09.359
It seemed like this week we were gonna have a quieter week in terms of big AI releases that you can actually use today.
00:00:09.519 --> 00:00:21.600
But then late Thursday, we got a new, soon-to-be open source model that completely reset the proprietary versus open framework.
00:00:21.920 --> 00:00:29.440
So, yeah, as always, some surprises, some new updates, and some quality of life improvements.
00:00:29.679 --> 00:00:44.719
So if you spend most of your week inside of AI tools, the problem is under the hood and sometimes only on a random tweet, these companies release some big updates that change how all of these products work.
00:00:44.880 --> 00:00:47.119
And unless you spend hours, you're gonna miss those.
00:00:47.200 --> 00:00:50.640
But that's what we do on our Friday features show.
00:00:50.880 --> 00:00:58.719
So on today's show, you're gonna learn what open source model is again resetting the open versus proprietary landscape.
00:00:58.799 --> 00:01:02.799
You're gonna learn how to claude on desktop is getting a lot more useful.
00:01:02.960 --> 00:01:11.920
And you're gonna see why OpenAI made another big change to its new and very popular Chat GPT Work Desktop app.
00:01:12.319 --> 00:01:13.680
All right, let's get into it.
00:01:13.840 --> 00:01:14.959
Welcome to Everyday AI.
00:01:15.040 --> 00:01:17.040
My name's Jordan Wilson, and we do this every day.
00:01:17.200 --> 00:01:25.280
This is your daily live stream podcast and free daily newsletter helping business leaders like you and me keep up with the nonstop avalanche of AI updates.
00:01:25.439 --> 00:01:27.280
I tell you what matters, what doesn't.
00:01:27.439 --> 00:01:30.640
You use that information to be the smartest person in AI at your company.
00:01:30.799 --> 00:01:32.319
Everyone's like, wow, how'd you do this?
00:01:32.480 --> 00:01:38.319
And you're like, well, I went to everyday your everydayai.com and signed up for the free daily newsletter.
00:01:38.400 --> 00:01:41.840
And well, you listen on the podcast and uh maybe on the live stream.
00:01:41.920 --> 00:01:42.959
So thanks for doing that.
00:01:43.200 --> 00:01:48.319
And if you are looking for that AI news, make sure to check it out in today's newsletter.
00:01:48.640 --> 00:02:04.799
All right, let's get straight into it and talk about our first new update, which I'm like, bless up, it's about time because I don't know about you, but with all these desktop apps, I've been trying to use them just like never leaving.
00:02:04.879 --> 00:02:10.719
So when I go into, as an example, codec slash chat GPT work, I try to not leave, right?
00:02:10.879 --> 00:02:12.639
Really change how I do my work.
00:02:12.800 --> 00:02:18.800
And one of the big downsides with Claude Desktop is you really couldn't do that until this week.
00:02:18.960 --> 00:02:24.080
That's because Anthropic did finally add their new desktop browser.
00:02:24.159 --> 00:02:27.439
So yeah, bringing a browser inside of Claude Desktop.
00:02:27.520 --> 00:02:40.479
So here's what's new Anthropic added a built-in browser to the Claude Code Desktop app, which lets Claude open external websites and actually click, type, and interact inside of them while it works on your code.
00:02:40.719 --> 00:02:50.159
So Claude can now open whatever you need inside of the Claude app, whether that's your API docs, a bug tracker, dashboard, whatever it may be.
00:02:50.319 --> 00:02:53.919
So it's not just reading the pages inside of there, it can act on them.
00:02:54.080 --> 00:03:01.919
Uh, Anthropics says safety is layered in as classifiers review every click and keystroke Claude makes on external sites.
00:03:02.159 --> 00:03:07.360
And the first time Claude touches a new site, you get a permissions card to allow or deny it.
00:03:07.439 --> 00:03:09.680
Uh, every site needs its own approval.
00:03:09.840 --> 00:03:11.520
So that's one of the things for me.
00:03:11.599 --> 00:03:13.360
I'm like, uh, that's a downside.
00:03:13.439 --> 00:03:26.639
Uh, but overall, this is a big step forward for anthropic, putting it more in parody, uh, mainly with codec slash chat GPT work and also cursor, which I've always said were the top two in terms of super app.
00:03:26.960 --> 00:03:32.159
Uh so now uh enthropic claw desktop can finally enter the conversation and we'll see.
00:03:32.240 --> 00:03:42.800
Presumably, uh, we'll have Windows uh with their new uh super app Microsoft Copilot entering, we think it should be probably by next month.
00:03:43.120 --> 00:03:59.199
So uh who has access, anyone running the Claude Code or Matt, uh, you know, I just call it the Claude Desktop app because it's not just Claude Code, uh, but the Claude Desktop app on Mac in Windows, and you do need a paid plan to take advantage of these features.
00:03:59.439 --> 00:04:05.280
Um, so an important choice though that they made the profile uses a completely clean profile.
00:04:05.439 --> 00:04:09.199
So none of your saved logins, none of your history, uh, etc.
00:04:09.439 --> 00:04:22.800
And that's one of the big updates that actually OpenAI rolled out to their Chat GPT work uh last week as they unveiled the non-technical version of uh codecs called the ChatGPT work.
00:04:23.040 --> 00:04:26.240
Well, it can import all of your Chrome history, everything.
00:04:26.480 --> 00:04:31.759
So a much more robust browser on the codex slash chat GPT work side.
00:04:31.920 --> 00:04:42.240
However, at least even for me personally, the way that I've changed my work, this was one of the big things that I just like I wasn't really touching Claude Desktop anymore.
00:04:42.480 --> 00:04:44.399
Uh, at least not very much, right?
00:04:44.480 --> 00:04:49.920
I still use it for you know adversarial, uh kind of working with codecs, right?
00:04:50.000 --> 00:04:57.439
I do some more advanced things that I'll I'll use claude desktop for, like I, you know, running GPT-5.6 soul inside of Claude Desktop.
00:04:57.600 --> 00:05:09.120
Conversely, I'm running Fable inside of Codec slash chat GPT work, but I really wasn't using Claude Desktop as much as I was, maybe like three or four months ago, mainly because of all the updates that Codex has.
00:05:09.279 --> 00:05:13.040
But at least for me, this one brings it a little bit more back into the conversation.
00:05:13.199 --> 00:05:14.560
So who's gonna find this valuable?
00:05:14.639 --> 00:05:19.759
I think Deb's obviously already working inside Cloud Code, uh, security teams, uh, right.
00:05:19.920 --> 00:05:31.120
Or if you're a general knowledge worker and you've been doing some things inside of Cloud Code and you're like, wow, it would be nice to bring, you know, some additional work that I do on the web into the fold.
00:05:31.279 --> 00:05:33.839
Uh, I think that's gonna be a big one for you.
00:05:34.319 --> 00:05:37.600
All right, let's go to our next update.
00:05:37.759 --> 00:05:45.759
So this one is more of a quality of life one that I'm like, yes, uh, I need this, and I think most people will enjoy it.
00:05:46.079 --> 00:06:00.000
So OpenAI launched universal search in Chat GPT, which lets you search in one place for all of your past chats, projects, images, and documents, all from one search box in the sidebar.
00:06:00.079 --> 00:06:06.480
Uh, so you can filter by content type uh and clicking a request that opens the chat project or file directly.
00:06:06.639 --> 00:06:10.879
So, yeah, this just launched a couple of days ago, kind of an under the hood.
00:06:10.959 --> 00:06:21.120
But again, this is one that should hopefully bring a big quality of life for the uh you know hundreds of millions of weekly active chat GPT users.
00:06:21.360 --> 00:06:27.279
The good thing this rolls out to everyone, whether you have a free pan, a free plan, uh paid plan, etc.
00:06:27.439 --> 00:06:32.000
And it did already roll out uh on the web and iOS and Android.
00:06:32.079 --> 00:06:33.040
So that's the other good thing.
00:06:33.199 --> 00:06:35.199
This also works inside of ChatGPT.
00:06:35.439 --> 00:06:45.360
So if you're anything like me, uh you spend a lot of time, regardless of what platform you're using, trying to find old chats, especially on the web.
00:06:45.519 --> 00:06:52.720
The good thing is like ChatGPT work and codecs are great at this because one thread can read every single other thread and direct you.
00:06:52.879 --> 00:07:02.639
But at least on the web, this has always been something I've struggled with because sometimes chats, uh, you know, these searches will only search normal chats, or they might not search through your project.
00:07:02.800 --> 00:07:08.160
So it is good to have a single universal chat, just makes Chat GPT much more accessible.
00:07:08.399 --> 00:07:10.240
So why is it useful?
00:07:10.399 --> 00:07:14.800
Well, I mean, your Chat GPT history just stops becoming a graveyard, right?
00:07:14.879 --> 00:07:22.399
That analysis that you ran in March that you put a ton of time in, and maybe you use some of it, but there was a lot of it that you were like, hey, we could benefit from this.
00:07:22.480 --> 00:07:23.199
Where is it?
00:07:23.360 --> 00:07:28.160
Uh, you know, the image that you did last week, the the contract you uploaded in May, right?
00:07:28.319 --> 00:07:33.199
It's all findable in seconds versus having to scroll forever, right?
00:07:33.360 --> 00:07:35.759
I would have loved this like a year or so ago.
00:07:35.920 --> 00:07:51.519
There were actually a few instances, uh, both in Chat GPT and Claude, where I had a very important, uh, couple of very important chats that I worked on that I literally couldn't find because sometimes that uh, you know, searching will literally only search the thread name.
00:07:51.680 --> 00:08:05.759
So in multiple instances, I've had to go through, export all of my chats, import them in, use a thinking mode just to find uh, you know, a certain chat that was at least for me very valuable or very useful that I needed to repurpose or reuse.
00:08:05.920 --> 00:08:06.959
So uh this is good.
00:08:07.040 --> 00:08:09.519
And like I said, this is rolling out to anyone.
00:08:09.680 --> 00:08:11.279
And who's gonna find this useful?
00:08:11.360 --> 00:08:17.680
I mean, if you're a power user of Chat GPT, this is an instant just quality of life upgrade.
00:08:18.000 --> 00:08:25.439
All right, our next one, not everyone uses superhuman, but um, I was actually surprised.
00:08:25.519 --> 00:08:33.279
So, to do, you know, to help me plan for this show, I also see, you know, hey, in our newsletter, what are those stories that people are clicking on the most?
00:08:33.360 --> 00:08:34.480
And I was actually surprised.
00:08:34.559 --> 00:08:37.759
This is one of our more clicked-on stories this past week.
00:08:37.919 --> 00:08:42.879
So apparently a lot of you all use superhuman for your mail.
00:08:42.960 --> 00:08:47.679
So, well, they launched a helpful new feature if you are a superhuman male user.
00:08:47.759 --> 00:09:04.639
And if you don't know what superhuman is, well, it's kind of confusing now because it used to just be the uh, you know, the I guess minimal and fast email program, but now superhuman itself, uh, the name is you know okay, let me just explain this a little better.
00:09:04.879 --> 00:09:21.360
So Grammarly, you know, a popular uh writing tool that a lot of people used, technically acquired superhuman, but they applied the superhuman uh name to all of Grammarly's other tools.
00:09:21.519 --> 00:09:29.039
So it is a little confusing now, but uh here's what's new if you do use superhuman with their new auto drafts.
00:09:29.120 --> 00:09:37.120
So superhuman launched a new version of auto drafts, which writes replies to your emails in your voice before you even open your inbox.
00:09:37.360 --> 00:09:47.360
So it does draft automatically two things responses to messages waiting on you, and follow-ups for emails other people haven't answered yet.
00:09:47.519 --> 00:09:52.159
So you open your inbox and the replies are sitting there for your review and to hit send.
00:09:52.320 --> 00:09:55.200
So the actual new thing here is the engine running it.
00:09:55.279 --> 00:10:04.240
So it now runs on frontier models from Anthropic and OpenAI, replacing an old GPT 3.5 version.
00:10:04.399 --> 00:10:10.799
Yeah, apparently they were still using GPT 3.5, which is maybe why the results weren't that good.
00:10:10.960 --> 00:10:15.279
So that's why now uh the drafts can actually sound like you.
00:10:15.440 --> 00:10:19.600
So right now, this is for paid plans only on business and enterprise.
00:10:19.759 --> 00:10:27.279
So you can set it up on the desktop and it syncs to mobile, and it also drafts the drafts sync with Gmail and Outlook.
00:10:27.519 --> 00:10:29.360
Uh, so who's gonna find it useful?
00:10:29.519 --> 00:10:36.399
Well, obviously, if you're a paid superhuman user struggling to keep up with emails, this is a good quality of life upgrade.
00:10:36.559 --> 00:10:40.399
And you know, at first I was almost like, should I even include this?
00:10:40.480 --> 00:10:50.799
Because this is one of those things that you can very easily do inside of uh Claude, you can very easily do inside of Chat GPT Codex Chat GPT work.
00:10:51.039 --> 00:10:52.240
But I'm like, yeah, there.
00:10:52.480 --> 00:10:55.840
I mean, if you're using these tools, keep using them, right?
00:10:56.000 --> 00:11:05.679
Uh, but for me, it it is technically very easy to build this inside of like ChatGPT work or codex, and it can send them all there.
00:11:05.759 --> 00:11:10.399
So part of me was like, all of this already exists and probably platforms you're already using.
00:11:10.480 --> 00:11:15.360
But if you all already a power superhuman user, uh, this is for you.
00:11:15.519 --> 00:11:23.039
Speaking of app specific uh updates, this one I'll probably use because I'm a heavy Spotify user.
00:11:23.200 --> 00:11:26.639
And if you are too, this is one that you might enjoy.
00:11:26.879 --> 00:11:39.759
So Spotify launched Talk to Spotify, which lets you have an actual conversation with the app by voice or text to control your music, discover new stuff, and also ask about your own listening history.
00:11:39.919 --> 00:11:44.480
So the cool thing here is well, it's back and forth, not just one-shot commands.
00:11:44.639 --> 00:11:55.360
So you can ask for an artist, then keep steering, and you can say just hit his recent stuff or make it more upbeat, and then you can save and cue and follow right on from the conversation.
00:11:55.519 --> 00:11:57.840
Uh, so who has access right now?
00:11:57.919 --> 00:12:05.759
It is in beta, all right, but it's rolling out to people in the US, Ireland, and Sweden on iOS and Android.
00:12:05.840 --> 00:12:11.519
So, yeah, you do have to be a paid uh user in one of those countries, like I am.
00:12:11.600 --> 00:12:15.360
I haven't actually checked this one yet because this one just came out, but I'll be using this one.
00:12:15.600 --> 00:12:17.440
So uh, why is it useful?
00:12:17.600 --> 00:12:19.600
Well, the differentiator is your data.
00:12:19.679 --> 00:12:23.919
So Spotify knows your playlist, your favorite artists, and your repeat listens.
00:12:24.000 --> 00:12:27.279
So it answers things like when did I first listen to this song?
00:12:27.440 --> 00:12:31.600
And you know, no general chatbot, at least right now, can do that, right?
00:12:31.759 --> 00:12:39.759
Although Spotify does have some integrations with like ChatGPT as an example, it doesn't have that level of granular data.
00:12:39.919 --> 00:12:54.480
Um, so this one for me, and the reason why I think it's very helpful, if you have a paid Spotify account and multiple people in your family use it, maybe if you have kids, you have you know people with wildly different tastes in music, but everyone's using the same Spotify account, right?
00:12:54.559 --> 00:13:03.600
Like if you have uh, you know, an Alexa or you know, you know, Home Speaker, like that's what I do, and everyone's listening to all different kinds of music on my Spotify account.
00:13:03.679 --> 00:13:10.399
So, you know, Spotify used to have, you know, and they still do, it's it's these weekly playlists that update based on your history.
00:13:10.639 --> 00:13:14.240
So I used to love these playlists, they would update every Friday.
00:13:14.320 --> 00:13:24.799
I would literally used to open it every Friday and be like, oh my gosh, I can't wait for my, you know, the two playlists are called uh, I think like Discover Weekly and Release Radar based on like your uh listening history.
00:13:24.960 --> 00:13:27.759
And you know, so recently mine is just all off.
00:13:27.919 --> 00:13:35.679
So this is one I'm gonna go in there and try to kind of you know discover new music based on the genres that I actually listen to.
00:13:35.840 --> 00:13:46.000
So this one obviously maybe a little bit more on the personal side, uh, but I think it's a pretty big update, especially if you are a power Spotify user like I am.
00:13:46.480 --> 00:13:51.679
All right, our next one in this one, if you are missing Sora from OpenAI, right?
00:13:51.840 --> 00:14:07.279
Remember uh about five, five or so months ago, you know, OpenAI kind of announced that they were killing off all their side quests and really focusing, you know, on their core products, which in theory has been great, and the company is growing just exponentially.
00:14:07.440 --> 00:14:15.360
Uh, but one of the downsides is there was these fantastic products that are now just shelfware, and one of those was Sora.
00:14:15.600 --> 00:14:24.159
And one of the most popular features of Sora was being uh able to essentially import your own self, your own avatar, right?
00:14:24.240 --> 00:14:29.440
You would have to verify your face, and then you could create videos with yourself in them.
00:14:29.679 --> 00:14:34.480
Well, uh Gemini Omni has finally rolled that out.
00:14:34.639 --> 00:14:38.000
So this is technically a new feature in two different parts.
00:14:38.240 --> 00:14:47.919
So, first, Google launched personal avatars inside of Gemini, which lets you record your face and voice, then generate AI videos of yourself on demand.
00:14:48.080 --> 00:14:54.960
So you just drop your avatar into prompts by typing the plus at sign uh and then your username.
00:14:55.039 --> 00:15:05.120
So, as an example, you would type, you know, create a video of at me uh singing with an orchestra, and then your likeness becomes a reusable asset inside of Gemini.
00:15:05.519 --> 00:15:12.000
So, right now you do have to have a paid account and be in the US and 18 or older.
00:15:12.159 --> 00:15:16.960
So, yeah, all the you know, EU, UK, uh not available right now.
00:15:17.120 --> 00:15:23.120
So, uh, which is all you know, people are always like, Why doesn't you know the EU get this?
00:15:23.279 --> 00:15:30.559
Well, the European exclusion is almost always about you know regulatory caution around biometric likeness, right?
00:15:30.639 --> 00:15:40.799
That's why so many features, especially ones like this that are hyper-personalized uh for your own likeness or others, you know, don't usually roll out sometimes at all to the EU.
00:15:41.360 --> 00:15:48.240
Uh, so and and so it's not only inside of Gemini, but it's also uh inside of Google Vids.
00:15:48.559 --> 00:15:52.960
So here is what they said about using this new feature inside of Google Vids.
00:15:53.200 --> 00:16:00.799
They said today we're rolling out two new updates to Google Vids to make it easier than ever to create, edit, and personalize your videos.
00:16:01.039 --> 00:16:03.519
Gemini Omni and personal avatars.
00:16:03.679 --> 00:16:06.960
Gemini Omni takes the hard work out of the editing process.
00:16:07.120 --> 00:16:12.320
So generating and refining high-quality clips is as easy as writing a simple prompt.
00:16:12.399 --> 00:16:18.480
And with personal avatars, you have an entirely new way to star in videos without setting up a camera.
00:16:18.639 --> 00:16:37.519
So, yeah, if you missed Google's big Gemini announcement, uh Gemini Omni announcement back in May, uh essentially it is the new family of VO models, but just much more powerful where their Vio, you know, so VO3, VO3.1, we're technically just video models, right?
00:16:37.679 --> 00:16:39.840
Gemini Omni is much more than that.
00:16:40.000 --> 00:16:43.279
It produces video, but it is like a world model.
00:16:43.440 --> 00:16:49.679
And I think it's much better at editing scenes uh than any other platform out there.
00:16:49.759 --> 00:16:51.440
Uh, so this is a pretty big one.
00:16:51.519 --> 00:16:55.679
I think uh, you know, who has you know, why is it useful?
00:16:56.080 --> 00:17:04.240
Because this is like the talking, you know, so many talking head videos that just need like background B-roll, right?
00:17:04.319 --> 00:17:09.440
So if you're someone in your company that's whether it's for internal or external purposes, right?
00:17:09.599 --> 00:17:15.359
I wouldn't use these to actually do like a full avatar, like talking head video.
00:17:15.519 --> 00:17:16.319
That's not the thing.
00:17:16.400 --> 00:17:19.680
But if you or someone else, well, like me, right?
00:17:19.759 --> 00:17:22.079
I'm constantly talking head video right now.
00:17:22.160 --> 00:17:23.519
I am a talking head video.
00:17:23.599 --> 00:17:37.039
Uh, you know, I don't necessarily not part of my brand or the everyday AI brand necessarily, but when I'm yapping about all these things, I could very well instead put an avatar of me working on these things.
00:17:37.119 --> 00:17:44.640
So you don't just have to, you know, look at my uh face made for radio and just be like, all right, guy, you know, put something else on the screen.
00:17:44.720 --> 00:17:47.119
So, but I think there's a ton of great use cases for this.
00:17:47.200 --> 00:17:54.720
So if you are in LD, uh, you know, training, uh, marketing, content creation, show social media, anything like this.
00:17:54.960 --> 00:18:03.440
And if you have a CEO as an example that's hard to nail down and you are trying to get more video content out there or something like this, right?
00:18:03.599 --> 00:18:09.359
Again, if you go through the proper channels, do all the approvals, data security, privacy, all that good stuff.
00:18:09.440 --> 00:18:14.880
But now this is such a a weapon to have in your creative arsenal, right?
00:18:15.039 --> 00:18:29.200
I would have loved to have this like 18 years ago at one of my first jobs where I was creating a lot of video and usually it was just kind of talking head of the CEO, and I would try to get some of this more like B-roll stuff, and it was just sometimes hard.
00:18:29.440 --> 00:18:40.079
So I think if you are a creative, if you are trying to have a a stronger uh, you know, presence uh on social media for your company, but it's a little hard.
00:18:40.240 --> 00:18:40.720
This is great.
00:18:40.799 --> 00:18:46.160
So yeah, content creators, marketers, educators, um, you know, is is gonna be huge.
00:18:46.400 --> 00:18:54.559
But I think the simple framing of this is it's kind of like a personal version of Hey Jen inside of Gemini, right?
00:18:54.640 --> 00:19:00.480
So Avatar video just went from being kind of a specialist tool uh to now going mainstream.
00:19:00.720 --> 00:19:07.839
All right, speaking of mainstream, let's tackle our next update because this one is very mainstream.
00:19:08.079 --> 00:19:19.039
So uh the new Chat GPT Work Desktop app has a big update that I think a lot of Chat GPT power users are going to enjoy.
00:19:19.279 --> 00:19:26.559
All right, so uh I'm just gonna go ahead and read uh what Tebow, the head of product at OpenAI, posted.
00:19:26.720 --> 00:19:29.039
It's a little easier just to read it, and this is brand new.
00:19:29.119 --> 00:19:33.759
I've been playing around with it a little bit today, but yeah, it hasn't even been out a full day yet.
00:19:33.920 --> 00:19:41.519
So uh Tebo said, uh evening, we've got we've gotten lots of great feedback on the new Chat GPT desktop app.
00:19:41.680 --> 00:19:46.079
So the work app, uh, which we didn't get totally right on the first try.
00:19:46.319 --> 00:19:48.640
And as a result, we made some changes.
00:19:48.799 --> 00:19:55.119
Number one, chat GPT conversation history and projects are now visible in the sidebar.
00:19:55.279 --> 00:20:01.039
Also, your chat and work history now sync across web, mobile, and desktop.
00:20:01.119 --> 00:20:04.079
Uh, local task will stay on your computer.
00:20:04.240 --> 00:20:15.359
Uh, then you can now easily switch between chat and work modes inside chat GPT on desktop, which is now also consistent with how it shows up on web and mobile.
00:20:15.519 --> 00:20:18.480
Nothing is changing for users on codex mode.
00:20:18.559 --> 00:20:22.000
Uh, Tebow says it's still the OG and best at what it does.
00:20:22.319 --> 00:20:24.640
So, what does this mean?
00:20:24.880 --> 00:20:33.839
Well, to actually say what it means, and you know, Tebow did kind of say it there, he said, yeah, we made some, we didn't totally get it right on the first try.
00:20:34.000 --> 00:20:36.880
And that kind of explains the update.
00:20:37.119 --> 00:20:49.839
Um so long story short, if you missed this, I just did uh an episode on this on Wednesday where we went over uh Chat GPT work, what it is, all that good stuff.
00:20:49.920 --> 00:20:55.839
But chat GPT work is essentially the non-technical version of Codex, right?
00:20:56.000 --> 00:20:59.039
OpenAI's uh autonomous desktop agent.
00:20:59.200 --> 00:21:04.960
Uh, but there is a chat GPT work on the web, and there is a chat GPT work on the desktop.
00:21:05.200 --> 00:21:14.799
So the problem was it the chat GPT work didn't exactly sync up very well with what you were doing on Chat GPT on the web.
00:21:15.039 --> 00:21:19.039
It was kind of this separate pop-up that came up in the right hand corner.
00:21:19.200 --> 00:21:32.880
So your chats were kind of there, but it just wasn't really intuitive to use because you would have all of your uh essentially your tasks in your projects that lived or started with chat GPT workslash codecs on the left sidebar.
00:21:32.960 --> 00:21:39.920
And then your chat GPT history was kind of this orphan page on the right hand side that wasn't really attached to anything.
00:21:40.079 --> 00:21:48.720
And then the big thing that people were like is hey saying, like, hey, I love Chat GPT, but I run everything inside of projects, and projects did not sync.
00:21:48.880 --> 00:21:58.880
So essentially, OpenAI changed and technically fixed all of that because now not only do your projects sync to the desktop version of ChatGPT work, right?
00:21:58.960 --> 00:22:01.279
But they're also in Left hand sidebar.
00:22:01.359 --> 00:22:09.680
And the other good thing is from an aesthetic standpoint, now it does look the exact same as it does on the web, which I think is going to help people.
00:22:09.839 --> 00:22:12.160
And it looks the exact same as it did on mobile.
00:22:12.240 --> 00:22:19.200
So the first variation of this it looked very similar uh on the web and on the mobile app.
00:22:19.279 --> 00:22:24.160
But then when you open the chat GPT work app, it looked and functioned completely different.
00:22:24.319 --> 00:22:34.240
So this is a great update uh for people who are trying out uh ChatGPT work, or if you're like me, right, I've been using Codex since day one.
00:22:34.400 --> 00:22:48.240
And the good thing is, well, now you have that ChatGPT work experience, which is just a very similar version of Codex, and then you have your familiarity with all of your uh normal Chat GPT chats and your uh projects.
00:22:48.400 --> 00:22:53.519
So if you're looking at this, if if you scroll down, it's all going to be under your recents.
00:22:53.680 --> 00:23:06.000
So whether you uh start a new task inside of ChatGPT work that's not attached to a project, it will go to your recents, and that's also where all of your new chats inside of Chat GPT will live.
00:23:06.160 --> 00:23:16.640
So if you do start a new chat inside of chat GPT.com, you don't attach it to a project, and then you go to ChatGPT work on the desktop, it will be there, which is great in the same place under recent.
00:23:17.279 --> 00:23:22.559
All right, our last piece of AI news, and this is technically the biggest one.
00:23:22.640 --> 00:23:35.440
So, not just a new heavyweight on the soon to be open source, and I'll explain that, but this one is definitely resetting the open versus close model paradigm.
00:23:35.599 --> 00:23:42.160
Yeah, we all thought it was the uh GLM 5.2 that was gonna do this, not anymore.
00:23:42.319 --> 00:23:49.039
Get ready because you're gonna be hearing a lot, especially if you are an AI large language model dork like I am.
00:23:49.200 --> 00:23:58.079
A lot of the conversation, I'm guessing for the next month or two is gonna be set around Kimmy K3, and this has huge implications.
00:23:58.160 --> 00:24:01.279
So, let me first explain what it is, what's new.
00:24:01.519 --> 00:24:17.839
So, uh Chinese AI startup Moonshot released Kimi K3, a 2.8 trillion parameter model that is now the largest open model ever built with a native vision mode and 1 million token context window.
00:24:18.160 --> 00:24:31.519
So here's the thing it came in at third place overall on the artificial analysis intelligence index, not far behind Claude Fable 5 and GPT 5.6.
00:24:32.000 --> 00:24:36.880
So uh that alone is mind-boggling, right?
00:24:36.960 --> 00:24:42.319
Uh, because we've been seeing this race just go back and forth, back and forth.
00:24:42.480 --> 00:24:53.119
And we thought that when Anthropic released Mythos and Fable 5, that this was essentially a new category, a new tier that no one else was ever going to touch.
00:24:53.279 --> 00:25:12.720
So not only uh, you know, about a month later did GPT 5.6 enter that category and on many of the most important benchmarks pass Fable 5, but now we have Kimmy K3, a soon-to-be open model that has not only entered the conversation, but it is actually surpassed.
00:25:12.799 --> 00:25:20.640
Yes, an open model has surpassed well both Fable and GPT 5.6 on many important benchmarks.
00:25:21.279 --> 00:25:30.160
Uh, so uh the best open model in the world now sits at three behind the two closed flagships.
00:25:30.319 --> 00:25:42.160
And the important thing that is thrusting this all back into the conversation is well, Anthropic is supposed to be pulling Fable 5 from subscriptions this Sunday.
00:25:42.400 --> 00:25:55.920
So not only do they have this continued pressure uh from GPT 5.6, but now there is an open model that you is gonna be better, even if you're paying$200 a month like I am, you're not gonna get access to Fable 5.
00:25:56.000 --> 00:26:09.440
There are obviously rumors that maybe today or maybe uh early next week we might be seeing Opus five, and maybe that will even reset and be much closer to Fable level than currently Opus 4.8.
00:26:09.599 --> 00:26:14.640
All right, anyways, let's talk a little bit more about Kimi K3, who has access to all that stuff.
00:26:14.880 --> 00:26:22.319
One important thing to note though, right now it's not technically an open source model, but it will be soon.
00:26:22.400 --> 00:26:28.000
Uh, Moonshot did say that they will be releasing the weights, so it will be an open weights model.
00:26:28.160 --> 00:26:36.720
Uh, they just haven't released the weights yet, but it just came out literally hours ago, and everything else that Moonshot has released in the past has been open.
00:26:36.880 --> 00:26:41.039
So this is live today in the Kimi app on Kimi.com.
00:26:41.119 --> 00:26:47.119
That's K-I-M-I, uh, and also in the Kimi Work Desktop app, as well as through the API.
00:26:47.440 --> 00:26:53.920
So they did say that they will release the full model weights by July 27th.
00:26:54.000 --> 00:26:55.920
So that's within like a week and a half.
00:26:56.000 --> 00:26:59.440
So at that time, anyone can download and self-host this.
00:26:59.519 --> 00:27:03.359
Obviously, it is a large 2.8 trillion parameter model.
00:27:03.519 --> 00:27:12.319
So uh if if you think you're gonna post this on normal consumer hardware, no, unless you've spent like 20 or 30,000 on your setup.
00:27:12.480 --> 00:27:14.880
Uh, so this is more for enterprise companies.
00:27:15.039 --> 00:27:18.960
Well, if you have the uh the GPUs, you can do this.
00:27:19.200 --> 00:27:22.079
So who's gonna find this valuable?
00:27:22.240 --> 00:27:24.079
Actually, no, let's first talk about why it's useful.
00:27:24.240 --> 00:27:30.559
Well, from the benchmarks alone, what we're seeing is its specialty is in long autonomous work, uh, right?
00:27:30.640 --> 00:27:47.680
So they kind of shared two different case studies where K3 designed and verified a working chip in a single 48-hour unsupervised run, and it reproduced an Astrophysics, uh, sorry, astrophysics research result in about two hours that they say would have normally taken a researcher, one to two weeks.
00:27:47.759 --> 00:27:58.960
And then the benchmarks obviously shows that it beats many models tested, including Fable 5 on web research, uh automation, spreadsheets, document reading, and more.
00:27:59.279 --> 00:28:06.240
And the crazy thing to me is on arena, which goes head to head, uh blind taste tests.
00:28:06.319 --> 00:28:13.279
One of the things that anthropic has really kind of dominated this space is front-end design, right?
00:28:13.440 --> 00:28:20.640
So um on the design arena, GPT-5.6 uh surpassed anthropic's models, including Fable 5.
00:28:20.720 --> 00:28:28.160
And then on the LM arena, which is kind of like user preference, this model is now number one, Kimi K3.
00:28:28.559 --> 00:28:37.039
So it's interesting because that has always been one of uh, you know, anthropic's kind of niches that they've owned, like front-end design.
00:28:37.200 --> 00:28:44.160
And now on the two most important front-end design benchmarks, they are no longer number one in either.
00:28:44.319 --> 00:28:48.000
So I'm personally excited to see what anthropic is gonna cook up.
00:28:48.160 --> 00:28:58.960
Maybe we'll see this with the open five drop, but I would assume that their next model, maybe they've let their foot off the gas in terms of front-end design because they've been so far ahead of everyone for so long.
00:28:59.119 --> 00:29:11.039
So I do and would assume that Opus five is probably gonna surpass Fable Five, at least in those areas, because that's something I'm sure that Anthropic is gonna be feeling a lot of pressure on.
00:29:11.200 --> 00:29:13.440
Aside from you know, pulling Fable 5.
00:29:13.519 --> 00:29:15.039
We'll see if they extend it again.
00:29:15.119 --> 00:29:30.000
Uh, but on the subscription package between GPT 5.6 and uh now Kimi K3, uh, yeah, they're gonna have to maybe justify why people are gonna be subscribing, but a lot of people are pointing to that just means we're getting an Opus V here soon.
00:29:30.160 --> 00:29:32.880
So, who's gonna find this new Kimmy K3 valuable?
00:29:32.960 --> 00:29:39.440
Uh, enterprises and developers who want frontier adjacent agents without that frontier pricing or vendor lock-in.
00:29:39.680 --> 00:29:59.839
Uh, so I mean, the crazy thing is, I mean, this is literally now an open, soon-to-be open model from a now from a Chinese uh you know startup that now benches ahead of Opus 4.8, and it is in the same tier, although technically lower than Fable 5 and GPT-5.6 soul.
00:30:00.000 --> 00:30:05.359
All right, so an interesting one here resetting the open versus closed race.
00:30:05.519 --> 00:30:08.079
And you know what it means for all of us, y'all?
00:30:08.240 --> 00:30:18.960
We're gonna continue to get more and more models, better models, and hopefully at cheaper prices now that we have an open model, presumably pushing the cost down.
00:30:19.039 --> 00:30:23.039
So, this one here with Kimi K3, definitely more of an enterprise play.
00:30:23.119 --> 00:30:30.480
Like I said, this is not really for inner uh for consumer hardware, although the prices are also much cheaper via the API.
00:30:30.640 --> 00:30:35.200
All right, so that's a wrap, a lot new that we went over today.
00:30:35.359 --> 00:30:36.880
I hope this one was helpful.
00:30:36.960 --> 00:30:44.240
Like I said, we do this every single Friday, our Friday feature show, where we bring you the latest and the greatest of what's new that you can actually use today.
00:30:44.400 --> 00:30:51.279
I was actually kind of bummed because we also had from Google a Gemini Notebook that came out, but it wasn't available to all paid users.
00:30:51.440 --> 00:30:52.160
So that's the thing.
00:30:52.240 --> 00:30:54.799
We do this if you have a base paid account.
00:30:54.960 --> 00:30:58.640
These are the things that you can use today to grow your company and career.
00:30:58.799 --> 00:30:59.039
All right.
00:30:59.119 --> 00:31:03.920
If this was helpful, do me a favor, please subscribe on the podcast and go to your everydayai.com.
00:31:04.079 --> 00:31:05.519
Sign up for the free daily newsletter.
00:31:05.599 --> 00:31:06.319
Thanks for tuning in.
00:31:06.400 --> 00:31:09.680
We'll see you back Monday and every day after that for more everyday AI.
00:31:09.839 --> 00:31:10.720
Thanks, y'all.
