August 20, 2026 • 27 mins
We discuss the fast-changing AI model landscape and how users are increasingly choosing models based on cost, convenience, and specific tasks.
We compare major closed and open-source models, outline which tools work best for coding, writing, real-time use, and image generation, and discuss how memory and context affect enterprise and power-user workflows.
------
🔒 Check Out Our Sponsor: LEDGER AGENT STACK 🔒
https://developers.ledger.com/docs/ai-tools/overview/?utm_source=Audio&utm_medium=Podcasts&utm_campaign=Limitless
------
🌌 LIMITLESS HQ ⬇️
NEWSLETTER: https://limitlessft.substack.com/
FOLLOW ON X: https://x.com/LimitlessFT
SPOTIFY: https://open.spotify.com/show/5oV29YUL8AzzwXkxEXlRMQ
APPLE: https://podcasts.apple.com/us/podcast/limitless-podcast/id1813210890
RSS FEED: https://limitlessft.substack.com/
------
TIMESTAMPS
0:00 Frontier Model Overload
1:09 Open Source vs Closed
3:20 Ranking Model Intelligence
5:41 Pricing Changes Everything
9:43 Cheap Models and Compute
11:14 Google and Grok Use Cases
15:30 Best Subscriptions For Most
17:41 Picking Models by Task
19:43 Agents and Knowledge Work
22:27 What Comes Next
24:50 Apple’s AI Wildcard
26:03 Final Model Recommendations
------
RESOURCES
Josh: https://x.com/JoshKale
Ejaaz: https://x.com/cryptopunk7213
------
Not financial or tax advice. See our investment disclosures here:
https://www.bankless.com/disclosures
Josh works with Anthropic as a contractor. All views expressed are his own and do not represent Anthropic, its leadership, or its affiliates. Nothing in this episode is investment advice.
Listen
Watch
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Josh:
Over the last 90 days, Frontier Labs shipped 15 plus models. (00:00):
undefined
Josh: OpenAI shipped three, Anthropic shipped four, Google shipped three, (00:04):
undefined
Josh: even Meta and Elon Musk shipped Frontier models. The Chinese shipped a bunch as well. (00:07):
undefined
Josh: And so if you're listening to the show, you're probably wondering, (00:12):
undefined
Josh: which model should I be using right now? (00:14):
undefined
Josh: Most of you likely have a GPT or Claude subscription, but you're wondering, (00:16):
undefined
Josh: should I be using these different models? (00:20):
undefined
Josh: The truth is, the Frontier landscape of AI models, when it was originally thought (00:22):
undefined
Josh: to be one or two, has now expanded to hundreds and hundreds of models. (00:26):
undefined
Josh: In OpenRata alone, you can access 400 plus. And so on this episode, (00:30):
undefined
Josh: we're going to dig into which model you should use for what particular use and (00:34):
undefined
Josh: when makes the most sense. (00:38):
undefined
Josh: And you'll realize that the argument has shifted not from using the most intelligent (00:40):
undefined
Josh: model, but maybe using the most cheaper model or using the model that's specific for you. (00:44):
undefined
Ejaaz: You got options, baby. There's a lot going on in the AI space and we're going (00:48):
undefined
Ejaaz: to help navigate that space because there is a lot of options and it does get (00:51):
undefined
Ejaaz: overwhelming. I think between the two of us, we've probably touched every single (00:55):
undefined
Ejaaz: one of these, at least a couple of times. (00:59):
undefined
Josh: I have too many subscriptions, Josh. (01:01):
undefined
Ejaaz: So yeah, between our 75 different subscriptions, we have covered these. (01:02):
undefined
Ejaaz: We have some feedback about which one to use for when, which is best for which use cases. (01:05):
undefined
Ejaaz: And I guess to start, we have this like pretty helpful visual companion artifact (01:10):
undefined
Ejaaz: here that can walk through kind (01:13):
undefined
Ejaaz: of what we're thinking, how we think about this. The first is the split. (01:15):
undefined
Ejaaz: There are two distinct classes of model when it comes to considering which ones (01:18):
undefined
Ejaaz: to use. The first is open source. (01:23):
undefined
Ejaaz: These are Chinese models predominantly. Most of these are your Kimis, (01:25):
undefined
Ejaaz: your DeepSeaks, like all of those models are Chinese open models. (01:29):
undefined
Ejaaz: And then we have the actual US models that are all closed source. (01:32):
undefined
Ejaaz: There's a big difference between the two. And we could see if we scroll down (01:36):
undefined
Ejaaz: a little bit, the difference in usage between these two, because in June of (01:38):
undefined
Ejaaz: 2025, US Labs accounted for 70% of the tokens generated. And now they're down to 30%. (01:42):
undefined
Ejaaz: The economics have changed widely. So now that 30% is worth much more than 70%. (01:48):
undefined
Ejaaz: But it is interesting to note that there has been this kind of reinvigoration (01:54):
undefined
Ejaaz: of Chinese models over the past couple of months. Now, granted, (01:57):
undefined
Ejaaz: this is based on open router data. (01:59):
undefined
Ejaaz: Open router is a single model router instance. This is not reflective of the norm. (02:01):
undefined
Ejaaz: But it's just worth noting that some of these open source models are pretty (02:06):
undefined
Ejaaz: powerful for the task at hand. (02:09):
undefined
Josh: Yeah, I think no one can debate the fact that people have shifted a lot using (02:12):
undefined
Josh: these open source models. And it's for a variety of different reasons. (02:16):
undefined
Josh: People want to own their own data, they want to run it privately at home. (02:19):
undefined
Josh: But the biggest shift, the biggest reason has been because these models are (02:22):
undefined
Josh: a lot cheaper. And open router data you mentioned only really represents a very (02:26):
undefined
Josh: niche sector of software engineers that want to experiment with a bunch of these models. (02:32):
undefined
Josh: But even in the enterprise world, where OpenAI and Anthropic are pretty dominant (02:36):
undefined
Josh: with their own model share, they've been losing the token market share to enterprises (02:40):
undefined
Josh: that are trialing and testing different models to save tens to hundreds of millions of dollars as well. (02:44):
undefined
Josh: And when I look at the main reason why, it's not because the open models are (02:50):
undefined
Josh: the most intelligent. You'll see in a second as we go through the scorecard, (02:55):
undefined
Josh: GLM 5.3 from China, amazing model, not as good as Fable 5. You'll look at Kimi (02:58):
undefined
Josh: K 2.7, you'll think the same thing. (03:02):
undefined
Josh: Those models are good enough to do the bulk of your own work. (03:05):
undefined
Josh: It doesn't sound too much like AI slop. (03:08):
undefined
Josh: It actually just speaks to you normally. And the biggest advantage is it has (03:11):
undefined
Josh: fewer safeguards, which has its pros and cons, but it basically does the tasks that you ask it to. (03:14):
undefined
Josh: Now, overall, it's fine if we talk about a bunch of models on this show, (03:20):
undefined
Josh: but it's good to get an overview, essentially, of how intelligent or how effective these models are. (03:24):
undefined
Josh: And what we have on the screen here is something known as the Artificial Analysis (03:30):
undefined
Josh: Index or Intelligence Index. (03:34):
undefined
Josh: And this is basically the best benchmark to test the general intelligence of (03:36):
undefined
Josh: these different models. (03:40):
undefined
Josh: Now, it may come as no surprise to you, but the Claude models, (03:41):
undefined
Josh: Anthropix models, top this. We've got Opus 5 at the top, which is their most (03:45):
undefined
Josh: recent model launch. We've got Claude Fable 5. (03:49):
undefined
Josh: And then, surprising to me, Josh, but Grok 4.6 from Elon, it is just, (03:51):
undefined
Josh: it is fantastic how effectively SpaceX has been able to pivot from being way behind the model race. (03:56):
undefined
Josh: They fired their entire AI team, hired a bunch of new folk, acquired Cursor (04:03):
undefined
Josh: for $60 billion and put out a model in, I think it's like the last six weeks (04:08):
undefined
Josh: that was able to contend with the top. And the best part is he has all the GPUs (04:13):
undefined
Josh: and compute to train them. (04:17):
undefined
Josh: And then if you go further down the stack, you'll see the likes of OpenAI, (04:18):
undefined
Josh: GPT 5.6 Sol, and a bunch of other Chinese open source models. (04:21):
undefined
Ejaaz: The question is, what do you make of this? As someone who is just a user of (04:25):
undefined
Ejaaz: AI, what do you make of this? How do you navigate this? (04:29):
undefined
Josh: Let me trial this, Josh. Let me ask you this question. Of these models that (04:31):
undefined
Josh: you see on the screen here, which ones are you using and for what specifically? (04:35):
undefined
Ejaaz: It's funny because about 90% of the usage comes from the top two. (04:41):
undefined
Ejaaz: And then 10% comes from everything else. And this is kind of like what I was (04:45):
undefined
Ejaaz: going for. I'm actually curious, is it a similar thing for you? (04:50):
undefined
Ejaaz: Like what kind of models are you using? (04:52):
undefined
Josh: Yes, but okay. So I would say it's around, I'm going to say like 75, 25. (04:54):
undefined
Josh: And the reason why I have that slight adjustment is because I kind of use the (05:01):
undefined
Josh: AI model that's most convenient to me wherever I'm getting the information or (05:06):
undefined
Josh: the inclination to use an AI model. So X is a perfect example, right? I'm scrolling X. (05:10):
undefined
Josh: I read something from one of these genius AI researchers and I'm like, (05:15):
undefined
Josh: I have no idea what the hell this means. (05:18):
undefined
Josh: I tap the Ask Rock button. I'm on Google. I'm searching something, right? I'm on my laptop. (05:20):
undefined
Josh: I'm like, okay, I'll just speak to Gemini in this. So in that sense, (05:25):
undefined
Josh: wherever the AI is conveniently placed, I use it there. (05:29):
undefined
Josh: And I have a feeling when Apple releases like their new AI model in a couple (05:32):
undefined
Josh: of months or a couple of weeks, actually, I'll probably use it on my phone as (05:35):
undefined
Josh: well because the models just generally are good enough to answer my basic questions. (05:38):
undefined
Ejaaz: Yeah, okay. So we're looking at these charts and we're seeing, (05:42):
undefined
Ejaaz: okay, here's where the intelligence is. Here's where things get a little less (05:44):
undefined
Ejaaz: intelligent, but much cheaper. (05:47):
undefined
Ejaaz: And if you're a normal consumer, I have a feeling that cost isn't that big of (05:48):
undefined
Ejaaz: a deal because the cost differences are really at scale. (05:52):
undefined
Ejaaz: What a lot of these companies offer, what Anthropoc offers, what OpenAI offers, (05:55):
undefined
Ejaaz: what Grok and Gemini offer is just the $20 a month, $100 a month, $200 a month plan. (05:58):
undefined
Ejaaz: You could just pay a fixed amount and get access to a lot of usage of these models. (06:04):
undefined
Ejaaz: So if you are a general consumer who is able to spend $20 per month on a subscription, (06:08):
undefined
Ejaaz: you're probably better served going to one of the major iolabs. (06:14):
undefined
Ejaaz: The product is better. You're going to get more intelligence. (06:16):
undefined
Ejaaz: You're going to get a more coherent product. (06:18):
undefined
Ejaaz: I find that a lot of the people who I speak to a lot of the time in my experience, (06:20):
undefined
Ejaaz: the only real reason to use these open source Chinese models is if you're consuming (06:23):
undefined
Ejaaz: tremendously large amounts of tokens for, say, agentic tasks. (06:28):
undefined
Ejaaz: Like if you're running an open claw instance or any sort of claw with a lot (06:32):
undefined
Ejaaz: of automated workflows, a lot of agents, you may want to defer to some of these (06:35):
undefined
Ejaaz: cheaper, perhaps the Chinese open source models just to save on costs on the low end. (06:39):
undefined
Ejaaz: But when it comes to day to day use on the high end, I'd say probably 90 plus (06:42):
undefined
Ejaaz: percent of the people watching this are best served just getting a subscription (06:46):
undefined
Ejaaz: to your favorite ai company (06:49):
undefined
Ejaaz: Using that product. That's kind of how I use it. It's funny that the remaining (06:52):
undefined
Ejaaz: 10% of prompts that aren't done through Anthropic are done mostly through Grok. (06:55):
undefined
Ejaaz: Like you mentioned, when I'm on X 24 hours a day, it's a good companion because (07:00):
undefined
Ejaaz: it has access to that database. (07:04):
undefined
Ejaaz: And then the other one is actually ChashyBT because they have a really amazing (07:05):
undefined
Ejaaz: image generation model. (07:09):
undefined
Ejaaz: And I love creating memes or sending images to my friends who are just like (07:10):
undefined
Ejaaz: using that as the visual component. I find that really compelling. (07:14):
undefined
Ejaaz: So that's kind of my stack is like mostly Anthropic, sometimes Grok, (07:17):
undefined
Ejaaz: sometimes chat gpt particularly for the image gen i think when it comes to image (07:21):
undefined
Ejaaz: gen we used to use nano banana and gemini all the time that was like the top (07:24):
undefined
Ejaaz: dog i think ever since image gen 2.0 release from chat gpt (07:28):
undefined
Ejaaz: is very compelling what i like is that it does get actual thinking behind it (07:32):
undefined
Ejaaz: so previously when you asked to generate an image it would take your words at (07:37):
undefined
Ejaaz: face value generate the image now it actually does some inference it does some thinking (07:40):
undefined
Ejaaz: it kind of interprets your message and interprets what it's seeing in the image (07:44):
undefined
Ejaaz: and then creates a much better output. So that's kind of how I would think (07:48):
undefined
Ejaaz: of navigating this um like chinese models are certainly a big deal but unless (07:51):
undefined
Ejaaz: you are kind of advanced super user consuming a ton of tokens perhaps they're not that valuable (07:55):
undefined
Josh: I will speak from my own personal experience i (08:00):
undefined
Josh: use the clawed models pretty much all the time (08:04):
undefined
Josh: but i've started using opus 5 more than fable 5 (08:07):
undefined
Josh: and and this might sound like a silly reason but it's because it speaks to me (08:11):
undefined
Josh: like a normal human being i don't know what the personality dials are on opus 5 versus fable 5 but (08:15):
undefined
Josh: they need to do that to fable 5 pronto because fable 5 just waffles and speaks (08:21):
undefined
Josh: in like archaic prose and i'm just like listen just give me (08:25):
undefined
Josh: give me the information that i've asked for and do not give me anything else (08:28):
undefined
Josh: i don't want to have to read, (08:31):
undefined
Josh: an essay every time i use this right now grok 4.6 i've actually noticed the (08:32):
undefined
Josh: intelligently josh i don't know if you've tried this on x you probably do when (08:37):
undefined
Josh: you're like kind of like trying to figure It speaks to you super intelligently (08:41):
undefined
Josh: and there's less crass about it. (08:45):
undefined
Josh: If you remember the earlier versions of Grok used to kind of like throw in some (08:47):
undefined
Josh: kind of like questionable words, slang and phrases. (08:52):
undefined
Josh: And now it just speaks to you like an intelligent human being. (08:55):
undefined
Josh: And I think this is because Elon's like, you know, SpaceX is IPO'd. (08:57):
undefined
Josh: He's got Cursor, very professional researchers involved, and they're trying (09:01):
undefined
Josh: to target a more enterprise oriented audience. And he said before that Grok (09:04):
undefined
Josh: 4.7 is on its way to being released in a few weeks now. He's aiming for Grok (09:08):
undefined
Josh: 5 by the end of the year. That's literally only in a couple of months. (09:12):
undefined
Josh: He is really targeting the enterprise landscape because he's seen Anthropic and OpenAI go after it. (09:15):
undefined
Josh: Now, same as you, on the Chinese models, I am not using any of them. (09:20):
undefined
Josh: And I admit this might be my own bias because... (09:25):
undefined
Josh: Mainly, I'm like, I'm not sure, like, I want to go to the efforts of signing (09:28):
undefined
Josh: up on an account, using those models, and then being like, well, (09:32):
undefined
Josh: I don't know whether this is giving me the right information. (09:36):
undefined
Josh: I don't know. I just kind of have this weird trust thing with the US and American (09:38):
undefined
Josh: brands. And maybe that's like my own fault. (09:41):
undefined
Josh: And then the one thing that I'll point out across all these different models (09:44):
undefined
Josh: that we make the point on on the screen here is. (09:47):
undefined
Ejaaz: I think a bulk (09:50):
undefined
Josh: Of the companies, Anthropic and OpenAI especially, have focused on making their (09:51):
undefined
Josh: models really good for enterprises. But as I mentioned earlier, (09:57):
undefined
Josh: the shift has really happened to open source because... (10:00):
undefined
Josh: More available models are cheaper and more accessible to people to use. They can run privately. (10:03):
undefined
Josh: But Anthropic and OpenAI's response to this is, we'll just distill our main (10:09):
undefined
Josh: model and give you cheaper models. (10:13):
undefined
Josh: So that's what OpenAI has done, right? You've got GPT 5.6. There's three versions (10:15):
undefined
Josh: of it. You've got Sol, Terra, and Luna. Luna is, I think, slashed 80% of the (10:19):
undefined
Josh: price than it was before already a week ago, right? (10:23):
undefined
Josh: So now it's like competing very much with GLM 5.3. (10:26):
undefined
Josh: And you've got Anthropic doing the same. I think Opus 5 is like half the cost (10:29):
undefined
Josh: of Fable 5, whilst being as capable as Fable 5. (10:33):
undefined
Ejaaz: And they also kept the SANA pricing low too, which is the lowest layer. (10:36):
undefined
Ejaaz: So that's like, yeah, there's plenty of options there. (10:39):
undefined
Josh: Plenty of options. And I think this comes down to one thing, (10:42):
undefined
Josh: and I don't want to make this about compute, but I have to, Josh. (10:45):
undefined
Josh: I think whichever lab, whether you're Chinese, whether you're like a low tier (10:48):
undefined
Josh: US American lab, or whether you're the high tier labs, compute will dictate (10:52):
undefined
Josh: whether you can serve these bottles cheaper, which will basically decide or (10:56):
undefined
Josh: determine which customers use your product. (11:00):
undefined
Ejaaz: Yeah. And I mean, this is ultimately coming down to your use case. (11:02):
undefined
Ejaaz: Like a lot of people won't ever run into this problem because they will never (11:04):
undefined
Ejaaz: need this thing called an API key. (11:07):
undefined
Ejaaz: They'll never actually pay per token. They'll oftentimes just pay through the subscription. (11:08):
undefined
Ejaaz: And if you're using a subscription, you have a couple of options. (11:12):
undefined
Ejaaz: Maybe this is a good time to go through these subscriptions and what are the (11:14):
undefined
Ejaaz: offerings of everybody. We have Gemini, which perhaps we could start there because (11:17):
undefined
Ejaaz: I feel like we've been mean to Gemini recently. We haven't talked much about Google. (11:21):
undefined
Ejaaz: What's that one? Yeah, like that one. (11:25):
undefined
Ejaaz: Who's that? They're currently at Gemini 3.1 Pro, which is nowhere really near the frontier (11:27):
undefined
Ejaaz: um anymore unfortunately they do have nano banana pro which is like this fun cool image generation (11:33):
undefined
Ejaaz: process they have what i find interesting about google and if you are interested (11:39):
undefined
Ejaaz: in these tools it may be interesting for a subscription google has these like (11:42):
undefined
Ejaaz: weird edge case tools that have really interesting harnesses one of them is for music production (11:45):
undefined
Ejaaz: i remember we've used this on the show many times it's really good at generating (11:49):
undefined
Ejaaz: lyrics and generating music in a way that sounds very good (11:54):
undefined
Ejaaz: I've only had that experience on a Google product. I haven't been able to use that anywhere else. (11:57):
undefined
Josh: Lyria. (12:01):
undefined
Ejaaz: Lyria, that's right. Yes, Lyria, Project Lyria. (12:02):
undefined
Josh: Remember we created the jingle for the show? (12:04):
undefined
Ejaaz: Yeah, so good, dude. So like Gemini and Google is like good for the weird stuff. (12:06):
undefined
Ejaaz: Like it'll generate you pretty good images. (12:10):
undefined
Ejaaz: It'll generate you fun music. They have this other tool. (12:11):
undefined
Ejaaz: This is in like the Google Lab suite where it'll generate you marketing material. (12:14):
undefined
Ejaaz: So if you have a lot of material from your brand or you have a bunch of logos (12:17):
undefined
Ejaaz: and you want marketing material, kind of say your logo's printed on something (12:20):
undefined
Ejaaz: that's staged nicely or you want custom merchandise, It'll do a lot of that (12:25):
undefined
Ejaaz: for you. So it's fun for those kind of narrow use cases of where the lab's products lie. (12:28):
undefined
Ejaaz: Outside of that, I don't know, I can't really recommend it that much. (12:32):
undefined
Ejaaz: It's not the strongest, not close to the strongest. (12:35):
undefined
Ejaaz: I'd say probably on top of that, we have Grok, which is very quickly catching up. (12:38):
undefined
Ejaaz: If you spend a ton of time on X like we do, Grok is your best friend. (12:41):
undefined
Ejaaz: Grok has access to that data that is available in real time on X. (12:45):
undefined
Ejaaz: If you're looking for current up-to-date news on pretty much anything, Grok is your go-to. (12:49):
undefined
Ejaaz: It can sort and cite specific tweets from specific moments that are happening (12:54):
undefined
Ejaaz: in real time and just generally speaking like you mentioned details it's very (12:58):
undefined
Ejaaz: good at just communicating with you directly (13:01):
undefined
Ejaaz: there's not a lot of fluff it is a science and technical model it is direct to the point (13:03):
undefined
Ejaaz: it'll get vulgar with you if you want it's fun to play around with there is (13:07):
undefined
Ejaaz: the voice mode which is really fun that i've i'd say of all the voice modes (13:10):
undefined
Ejaaz: that's the voice mode i've had the most fun with is because it's just so (13:14):
undefined
Ejaaz: ridiculous and oftentimes if you see memes online it is because is the voice behind it. (13:18):
undefined
Josh: The GPT voice mode just jars me so much. I don't know, have you spoken to it. (13:22):
undefined
Ejaaz: With the recent update? The new version, I have, yes. So when you say that, what do you mean? (13:27):
undefined
Josh: And so it sounds more human, right? It like pauses, but it sounds like it's (13:30):
undefined
Josh: like some sassy person that's talking to me. It does. Like I asked this question and then he goes, (13:35):
undefined
Josh: hmm, yeah, yeah, like, yeah, like, listen, like, I get it. (13:41):
undefined
Josh: And, you know, this is what the weather is. And I'm like, yo, (13:45):
undefined
Josh: I don't need this. I just asked you what the weather is. (13:49):
undefined
Josh: Like, you know, just tell me what's up. But I agree with you largely on like on Google and SpaceX. (13:51):
undefined
Josh: What I will say about Google, Josh, is I think they have a similar profile to (13:56):
undefined
Josh: Microsoft. I'm interested if you agree with me here. (14:02):
undefined
Josh: Microsoft is not a name that I would draw against the top frontier AI Model (14:05):
undefined
Josh: Labs. In fact, I think they fumbled the bag massively. (14:10):
undefined
Josh: But they are embedded across pretty much every single Fortune 500 company, (14:13):
undefined
Josh: whether we like it or not. Like we just live in our tech bubble. (14:17):
undefined
Josh: But outside of the tech bubble, people run Microsoft Teams and the Microsoft Suite. (14:20):
undefined
Josh: And I think Google's in a similar position, essentially, like a lot of new upstarts (14:24):
undefined
Josh: use Google Docs, Google Suite, and they're just going to click the Gemini button (14:28):
undefined
Josh: in the same way that I do so because it's just convenient. Do you agree with that? (14:31):
undefined
Ejaaz: Yeah, I guess there's just a moat to being like the biggest company in the world (14:35):
undefined
Ejaaz: and having access to all of these kind of legacy enterprise companies. (14:38):
undefined
Ejaaz: It feels like anyone who has built a business over the last 20 or 30 years has (14:42):
undefined
Ejaaz: been using Microsoft, has been using Google. (14:45):
undefined
Ejaaz: They're going to continue to leverage that. It seems like Google is slightly (14:47):
undefined
Ejaaz: more ahead than Microsoft, if I had to guess. I'm much more excited about Google (14:51):
undefined
Ejaaz: as a company than Microsoft is, but they both benefit from that legacy customer base. (14:55):
undefined
Josh: Well, both are amazing VCs, right? (14:59):
undefined
Ejaaz: Both are incredible VCs. Yes. I think they should be judged not on the quality (15:02):
undefined
Ejaaz: of their product, but the quality of their investment in the other competitors (15:05):
undefined
Ejaaz: that are going to crush them. (15:08):
undefined
Ejaaz: But I do like that, right? It's like they have a hedge. Microsoft owns a large (15:11):
undefined
Ejaaz: part of OpenAI. That's incredible. (15:16):
undefined
Ejaaz: Google owns this like massive stake in SpaceX, in Anthropic, (15:18):
undefined
Ejaaz: in both of these companies that are going to like be huge IPOs, (15:21):
undefined
Ejaaz: one of which already did. (15:25):
undefined
Ejaaz: So they are amazing venture funds, perhaps slightly less better AI labs. (15:26):
undefined
Ejaaz: The two that don't have to benefit from this kind of legacy software are the (15:30):
undefined
Ejaaz: two that we're probably going to recommend. (15:35):
undefined
Ejaaz: And in fact, if you scroll down a little bit further, we could see the box of (15:37):
undefined
Ejaaz: price against intelligence and kind of where each of these models sit. (15:41):
undefined
Ejaaz: Yeah, this chart right here. (15:44):
undefined
Ejaaz: And we're looking at those in the top right. this is where most of the people (15:46):
undefined
Ejaaz: are going to want to live this is where i think we spend all of our time you're (15:49):
undefined
Ejaaz: either getting a chat gpt or again a cloud membership and like that's that's (15:52):
undefined
Ejaaz: for the most part that's what most people need uh the differences are (15:55):
undefined
Ejaaz: small but noteworthy mostly as it relates to the models i mean both of them (15:59):
undefined
Ejaaz: i can't recommend enough like spend the 20 bucks a month try it just try it (16:04):
undefined
Ejaaz: for a month if you don't have it see what you think if you run out of tokens (16:07):
undefined
Ejaaz: upgrade to 100 a month this is access to the same intelligence that all these (16:10):
undefined
Ejaaz: researchers are making math breakthroughs with (16:14):
undefined
Ejaaz: and it's really impressive and really capable now each (16:16):
undefined
Ejaaz: one of these is slightly different i'd say claude is a little more tailored towards coding (16:20):
undefined
Ejaaz: writing and it's a really strong general purpose model i find with gpt 5.6 soul (16:26):
undefined
Ejaaz: in particular it's very good at being the (16:32):
undefined
Ejaaz: High level operator so if you're working on like complicated tasks it's good (16:35):
undefined
Ejaaz: at creating spec sheets it's good at running checks against the things that (16:39):
undefined
Ejaaz: you do and then oftentimes like i find that that claude and the fabled models (16:41):
undefined
Ejaaz: and the opus models are very good at implementation (16:45):
undefined
Ejaaz: and they're very good with less guidance so the way i've been using these models (16:47):
undefined
Ejaaz: recently is with less and less guidance i think earlier on (16:50):
undefined
Ejaaz: i had this whole skill sheet and it was like 15 different skills that would (16:55):
undefined
Ejaaz: automate different things for me and slowly over time that skill sheet has gone (16:58):
undefined
Ejaaz: lower and lower and lower and we were talking actually before the show (17:01):
undefined
Ejaaz: the best way to extract the most value out of these models is just to (17:04):
undefined
Ejaaz: get out of the way give it the end goal and say like hey go do this thing for (17:07):
undefined
Ejaaz: me i trust that you have the intelligence to do this i'm not going to put guardrails (17:12):
undefined
Ejaaz: on it you have the context you can go and do this for me and i think that's (17:14):
undefined
Ejaaz: mostly where i find myself using this personally is i use fable (17:17):
undefined
Ejaaz: i i do not leave any fable tokens unused very valuable tokens each week and (17:21):
undefined
Ejaaz: then opus is the fallback for the general workhorse model and it's been a really (17:25):
undefined
Ejaaz: powerful combo of just kind of (17:28):
undefined
Ejaaz: being able to accomplish anything that you want it connects to all of my (17:30):
undefined
Ejaaz: services i have it connected to my email to my google drive it has access to (17:33):
undefined
Ejaaz: my files the folders the context (17:36):
undefined
Ejaaz: and it's just a really helpful all-in assistant it works really well. (17:38):
undefined
Josh: I think people are also probably wondering, okay, well, what if I don't really (17:42):
undefined
Josh: care about the general intelligence as much? (17:45):
undefined
Josh: What if I'm trying to do a specific thing? Which model should I use then? (17:47):
undefined
Josh: I think there are two things to consider here. Number one is, (17:50):
undefined
Josh: if you are a software engineer, like AI models have been all the rage for their (17:54):
undefined
Josh: coding capabilities, but maybe a bunch of you actually like code hardcore companies (17:58):
undefined
Josh: and you want to figure out which model you should use. (18:03):
undefined
Josh: I would say in that question, the clawed models and the GPT 516 models are pretty (18:05):
undefined
Josh: high up there. Codex usage has gone from 5 million users to 15 million users (18:12):
undefined
Josh: in about a month and a half. (18:17):
undefined
Josh: I am tired of seeing Thibaut, who is the head of Codex at OpenAI, (18:18):
undefined
Josh: tweets on my timeline every single time he gets a million user update. (18:22):
undefined
Josh: But these models are very powerful at not only understanding and reading your (18:26):
undefined
Josh: code base, but intuitively figuring out what product or feature you should build next. (18:30):
undefined
Josh: I have a ton of feedback from friends that work at companies that are tech adjacent (18:35):
undefined
Josh: and maybe not even in tech at all, which use these models to build their premium (18:40):
undefined
Josh: features. The other thing I'll say is. (18:45):
undefined
Josh: You know, we're talking about the top right box over here, which is essentially (18:48):
undefined
Josh: the Pareto Frontier. So if you had to sacrifice something of cost or intelligence (18:51):
undefined
Josh: or blah, blah, blah, you would still be using these types of models. (18:56):
undefined
Josh: But if I had to take a bet, if I was a betting man, I would say Grok 4.6. (18:59):
undefined
Josh: Quen 3.8, and Kimi will be inside this box within a couple of months time. (19:04):
undefined
Josh: That's going to change the way we use these different types of models. (19:09):
undefined
Josh: I said before we started this show, it's unsexy to say. (19:12):
undefined
Josh: But I think the number one AI company that will come out in the next 12 months (19:16):
undefined
Josh: will be some form of aggregator platform. We saw that Stripe just acquired Open (19:20):
undefined
Josh: Router for $7 billion, allegedly. (19:24):
undefined
Josh: We put out an episode of this yesterday. Definitely go check that out. (19:27):
undefined
Josh: But I think we're going to see these platforms that help you pick and choose (19:30):
undefined
Josh: which models to use at the right time. (19:34):
undefined
Josh: It aggregates your memory so it already knows what you want to do. (19:36):
undefined
Josh: And so you don't have any of the complications around that. I think we'll see (19:38):
undefined
Josh: a bunch of these models kind of step up there. (19:41):
undefined
Josh: Now, aside from coding, if you aren't a coder, but if you are, (19:43):
undefined
Josh: let's say, a general knowledge worker, you go to work, you use email, (19:47):
undefined
Josh: you use Slack, you use a bunch of these other plugins that you also mentioned. (19:50):
undefined
Josh: There are models that are specifically good for agentic tool use. (19:53):
undefined
Josh: And actually, my most recent favorite is Grok 4.6 or GrokBot specifically. (19:57):
undefined
Josh: This was released from Elon Musk and SpaceX, I think last week. (20:02):
undefined
Josh: And it basically is an agent or multiple agents that spins up in a virtual machine (20:06):
undefined
Josh: on like in the cloud. So you don't have to worry about it running on your laptop (20:11):
undefined
Josh: and hacking all your stuff. (20:14):
undefined
Josh: And you can give it access to any and every tool and it intuitively understands (20:15):
undefined
Josh: and gets what you want to do. (20:19):
undefined
Josh: It can learn what you do over time and it improves. (20:20):
undefined
Josh: So I think when you look at the Metamuse Spark 1.2 and you might think, (20:23):
undefined
Josh: oh, that's a meta model. I don't ever use meta. (20:27):
undefined
Josh: These types of models are gonna become more available for general usage or knowledge work. (20:30):
undefined
Josh: And I think if that's something that you're inclined to use and you don't really (20:34):
undefined
Josh: care about the Google search stuff, you'll use Claude for that anyway. (20:36):
undefined
Josh: These are models that I'll probably look into because they are superior in many ways. (20:39):
undefined
Ejaaz: Yeah, it's fun to pick one and stick with it. I find that like oftentimes with (20:43):
undefined
Ejaaz: the aggregators, the right time to use that if you're doing lots of work, (20:46):
undefined
Ejaaz: if you're generating lots of tokens, if you are just a person who wants to use (20:49):
undefined
Ejaaz: AI for their day-to-day tasks to help you come up with like a grocery list, to help you (20:52):
undefined
Ejaaz: cook specific things, to help you go to the gym and give you workout classes and ideas. (20:56):
undefined
Ejaaz: It's helpful to pick a singular model, a singular service, and then just go deep with that. (21:01):
undefined
Ejaaz: A lot of the difference makers at this point because i mean most people aren't (21:07):
undefined
Ejaaz: using these models for frontier intelligence they're not going to solve novel (21:10):
undefined
Ejaaz: math problems they're figuring out what time they need to get to like the store (21:13):
undefined
Ejaaz: or the school to like pick up their kids and they just need some help they need a helpful assistant (21:17):
undefined
Ejaaz: the most helpful thing is context because all these models are more than capable (21:22):
undefined
Ejaaz: of those kind of lower level tasks the difference maker is the context that (21:26):
undefined
Ejaaz: it knows about you it needs to understand what are your dietary preferences (21:29):
undefined
Ejaaz: What are you or anyone in your family allergic to? (21:32):
undefined
Ejaaz: What have you made in the past that went well, that didn't go well? (21:34):
undefined
Ejaaz: And it kind of collects this database of information about you that allows it (21:38):
undefined
Ejaaz: to make better decisions going forward. (21:41):
undefined
Ejaaz: And that, at the end of the day, is ultimately what the difference maker is. (21:43):
undefined
Ejaaz: For me, at least, when deciding what model to use, it's like, (21:47):
undefined
Ejaaz: okay, which model has all the context (21:49):
undefined
Ejaaz: about me that can help solve my very specific task that I have here? (21:51):
undefined
Ejaaz: Oftentimes, the answer is Claude because I've been working with it for so long. (21:55):
undefined
Ejaaz: It just has this, like, huge chain of context. So whenever I ask, (21:58):
undefined
Ejaaz: like, hey, I need some help in the gym this week, it knows what I've been up (22:01):
undefined
Ejaaz: to, it knows where I'm at, it knows what the weight has been, (22:04):
undefined
Ejaaz: it knows what the food has been, and it's able to kind of curate this very custom stack against that. (22:07):
undefined
Ejaaz: And as someone who is just, you know, not really building anything crazy, (22:11):
undefined
Ejaaz: they're not using millions of tokens a week, they're just looking for an AI companion. (22:14):
undefined
Ejaaz: That's kind of how you can think about it. It's like, get a membership, (22:18):
undefined
Ejaaz: try it out, feed it a bunch of context about yourself, and then get your own (22:20):
undefined
Ejaaz: personal assistant. And that's kind of, I think, like the best route for most people. (22:24):
undefined
Josh: So I think to wrap up this episode, (22:27):
undefined
Josh: It's okay talking about, you know, what the frontier landscape looks like today. (22:30):
undefined
Josh: But the question is, what is it going to look like in a few months from now? (22:33):
undefined
Josh: And I say a few months specifically because this stuff moves too quickly. (22:37):
undefined
Josh: I saw an update from Sam Altman at OpenAI yesterday where they announced that (22:41):
undefined
Josh: they are slowing down some of their model training runs. (22:46):
undefined
Josh: And that's because they've noticed that a bunch of internal, (22:51):
undefined
Josh: more intelligent, unreleased models that they have built has become a lot more (22:53):
undefined
Josh: misaligned than previous models, which means that it could pose as a threat (22:59):
undefined
Josh: or danger to any user who gets access to it. (23:02):
undefined
Josh: And I've noticed the same similarly maybe from Anthropic and a bunch of other (23:06):
undefined
Josh: frontier labs, where we're starting to see a little bit of a slowdown. (23:09):
undefined
Josh: And slowdown isn't in the sense that they're necessarily stopping training full (23:14):
undefined
Josh: stop. Obviously, they're still training internal models. (23:17):
undefined
Josh: They might be likely to not release models or more intelligent models going (23:20):
undefined
Josh: forward because of government regulation. (23:24):
undefined
Josh: And I think this is going to allow a bunch of other Frontier Labs that are behind (23:27):
undefined
Josh: in this race to be able to catch up. (23:31):
undefined
Josh: So if I had to guess what three months from now, let's say at the end of the year, right? (23:32):
undefined
Josh: If I were to make a prediction, I think we're going to have about three to five (23:37):
undefined
Josh: really good open source models that are as capable as the smartest model that (23:41):
undefined
Josh: you have access to today from the Frontier Labs like Anthropic and GPT. (23:46):
undefined
Josh: And I think they're going to have fewer safeguards. So you can use it for any (23:50):
undefined
Josh: and every use case. I think you can run it privately at home, (23:53):
undefined
Josh: maybe even off of your laptop. So that changes dynamics quite a lot. (23:56):
undefined
Josh: I think we're going to have Apple entering the game with their own locally run (23:59):
undefined
Josh: and trained AI model, which I think is going to change the game because everyone, (24:04):
undefined
Josh: 3.5 billion people in the world right now have one of these or one of the Apple devices. (24:08):
undefined
Josh: They're going to run that locally. I think it's going to look quite different. (24:12):
undefined
Josh: I'm curious, you know, We started off this episode with the token split. (24:15):
undefined
Josh: I wonder what that's going to look like three months down the line. (24:19):
undefined
Josh: And then the last thing I'll say is, and this might be from my background in (24:22):
undefined
Josh: general, but I think once someone or. (24:25):
undefined
Josh: A bunch of companies make it easier to run models locally at home, (24:28):
undefined
Josh: I think people are going to play around with that more because it allows you to (24:33):
undefined
Josh: connect it to your fitness app data and not share too many personal anecdotes (24:36):
undefined
Josh: or data profiles with Frontier Labs, which again, they can use to train their own models. (24:41):
undefined
Josh: You may not want to hand that over, And so I think we'll see a rise of open (24:45):
undefined
Josh: source models. That's just my guess. That's interesting. (24:49):
undefined
Ejaaz: Okay, I'm taking the other side. I'm thinking that no one's going to go through (24:51):
undefined
Ejaaz: the trouble of downloading the weights and running their own open source models, (24:54):
undefined
Ejaaz: that Apple is just going to own that entire world. (24:57):
undefined
Ejaaz: That like for all of the people in the United States that own Apple devices, (25:00):
undefined
Ejaaz: they're just like, no one's going to even have an idea that they're using AI. (25:03):
undefined
Ejaaz: It's just going to be Siri. It's going to be competent. It's going to be better. (25:06):
undefined
Ejaaz: It's already going to be pre-downloaded. I think the user experience is really (25:08):
undefined
Ejaaz: important. And open source has a pretty horrific user experience. (25:11):
undefined
Ejaaz: You have to download, run the weights. you often need a lot of hardware to do that. (25:14):
undefined
Josh: But I'm counting Apple as locally run at home because it's encrypted. (25:17):
undefined
Ejaaz: Right? Well, if you're counting Apple, (25:20):
undefined
Ejaaz: sign me up because um we got a lot coming from them their event is happening (25:22):
undefined
Ejaaz: in like two weeks or something it's very soon we see (25:26):
undefined
Josh: Some airports with some cameras you sent me a video. (25:28):
undefined
Ejaaz: Yeah dude we've been getting lots of leaks lately it's really good maybe we have to have cameras (25:31):
undefined
Ejaaz: yes yes it's gonna be visuals and then they're gonna have apple intelligence (25:36):
undefined
Ejaaz: baked into it and it's like oh maybe we need a leak episode prior to the actual (25:39):
undefined
Ejaaz: new iphone unveiling because we have basically now the entire checklist of all (25:43):
undefined
Ejaaz: the things that are going to be revealed and (25:46):
undefined
Ejaaz: this to me as like apple fanboy plus ai fanboy is going to be the biggest event (25:48):
undefined
Ejaaz: ever pretty much because this is the actual rollout of the ai that they've been (25:53):
undefined
Ejaaz: promising us and failing to deliver for so long mixed with this brand new suite (25:56):
undefined
Ejaaz: of products that we've never seen before allegedly (26:00):
undefined
Josh: Up to the delayed two years. (26:02):
undefined
Ejaaz: But for those of you who came here for models that is the model update that's (26:04):
undefined
Ejaaz: just about when you'd want to (26:07):
undefined
Ejaaz: use each one like if you're interested in being on x real-time news feed (26:08):
undefined
Ejaaz: you want to go with grok if you like to yap a lot the voice model on chat gpt (26:12):
undefined
Ejaaz: is pretty exceptional that might be a good place for you if you like to be a (26:15):
undefined
Ejaaz: little more intellectual be thoughtful if you want really just the bleeding-edge (26:19):
undefined
Ejaaz: models the highest intelligence go with claude (26:21):
undefined
Ejaaz: and if you like making music and doing just like fun cute things with ai gemini (26:23):
undefined
Ejaaz: is actually kind of a compelling product (26:27):
undefined
Ejaaz: there's something for everybody that is the general overview of the models i (26:29):
undefined
Ejaaz: hope you enjoyed this is going to change certainly in the next month or two (26:32):
undefined
Ejaaz: whenever these new models i mean we have (26:36):
undefined
Ejaaz: the new astro model from open ai that has been teased for the last eternity (26:37):
undefined
Ejaaz: it seems like it's been held up it's going through some i don't know safeguard (26:41):
undefined
Ejaaz: governmental checks but that's coming (26:44):
undefined
Ejaaz: so this is set to change tbd but for now that is the state of the model address (26:46):
undefined
Ejaaz: and yeah hope you guys enjoyed watching (26:50):
undefined
Josh: Yeah and i'm curious for those of you who are listening what do we miss on this (26:52):
undefined
Josh: episode? Are you using models in a very different way? (26:57):
undefined
Ejaaz: Are you using meta models? (26:59):
undefined
Josh: Yeah, is anybody using meta models? Are there any Facebook users out there that (27:01):
undefined
Josh: are using meta AI intelligence? (27:05):
undefined
Ejaaz: Please fill me in. (27:06):
undefined
Josh: I will say, if you are a marketer or an advertiser, you're probably using meta's (27:07):
undefined
Josh: model and it's probably making you a hell of a ton more money. (27:11):
undefined
Josh: If that is you, let us know. If there are any other use cases that we haven't (27:14):
undefined
Josh: mentioned, please let us know. If you are a locally run open source fan and (27:17):
undefined
Josh: you're saying, no, people will download the weights, tell us why. (27:21):
undefined
Josh: Let us know in the comments, DM us. We read any and every single message. (27:24):
undefined
Josh: Now, if you're listening to this or watching this on YouTube, (27:28):
undefined
Josh: Spotify, Apple, or wherever you're listening to this, please subscribe. (27:30):
undefined
Josh: Please turn on notifications and leave us a comment. It helps us out massively. (27:33):
undefined
Josh: And share it with a friend as well. It helps us out. And I think that is pretty (27:37):
undefined
Josh: much it. Thank you so much for listening. And we will see you on the Roundup. (27:41):
undefined
Ejaaz: See you on the Roundup. (27:45):
undefined
undefined
Josh: OpenAI shipped three, Anthropic shipped four, Google shipped three, (00:04):
undefined
Josh: even Meta and Elon Musk shipped Frontier models. The Chinese shipped a bunch as well. (00:07):
undefined
Josh: And so if you're listening to the show, you're probably wondering, (00:12):
undefined
Josh: which model should I be using right now? (00:14):
undefined
Josh: Most of you likely have a GPT or Claude subscription, but you're wondering, (00:16):
undefined
Josh: should I be using these different models? (00:20):
undefined
Josh: The truth is, the Frontier landscape of AI models, when it was originally thought (00:22):
undefined
Josh: to be one or two, has now expanded to hundreds and hundreds of models. (00:26):
undefined
Josh: In OpenRata alone, you can access 400 plus. And so on this episode, (00:30):
undefined
Josh: we're going to dig into which model you should use for what particular use and (00:34):
undefined
Josh: when makes the most sense. (00:38):
undefined
Josh: And you'll realize that the argument has shifted not from using the most intelligent (00:40):
undefined
Josh: model, but maybe using the most cheaper model or using the model that's specific for you. (00:44):
undefined
Ejaaz: You got options, baby. There's a lot going on in the AI space and we're going (00:48):
undefined
Ejaaz: to help navigate that space because there is a lot of options and it does get (00:51):
undefined
Ejaaz: overwhelming. I think between the two of us, we've probably touched every single (00:55):
undefined
Ejaaz: one of these, at least a couple of times. (00:59):
undefined
Josh: I have too many subscriptions, Josh. (01:01):
undefined
Ejaaz: So yeah, between our 75 different subscriptions, we have covered these. (01:02):
undefined
Ejaaz: We have some feedback about which one to use for when, which is best for which use cases. (01:05):
undefined
Ejaaz: And I guess to start, we have this like pretty helpful visual companion artifact (01:10):
undefined
Ejaaz: here that can walk through kind (01:13):
undefined
Ejaaz: of what we're thinking, how we think about this. The first is the split. (01:15):
undefined
Ejaaz: There are two distinct classes of model when it comes to considering which ones (01:18):
undefined
Ejaaz: to use. The first is open source. (01:23):
undefined
Ejaaz: These are Chinese models predominantly. Most of these are your Kimis, (01:25):
undefined
Ejaaz: your DeepSeaks, like all of those models are Chinese open models. (01:29):
undefined
Ejaaz: And then we have the actual US models that are all closed source. (01:32):
undefined
Ejaaz: There's a big difference between the two. And we could see if we scroll down (01:36):
undefined
Ejaaz: a little bit, the difference in usage between these two, because in June of (01:38):
undefined
Ejaaz: 2025, US Labs accounted for 70% of the tokens generated. And now they're down to 30%. (01:42):
undefined
Ejaaz: The economics have changed widely. So now that 30% is worth much more than 70%. (01:48):
undefined
Ejaaz: But it is interesting to note that there has been this kind of reinvigoration (01:54):
undefined
Ejaaz: of Chinese models over the past couple of months. Now, granted, (01:57):
undefined
Ejaaz: this is based on open router data. (01:59):
undefined
Ejaaz: Open router is a single model router instance. This is not reflective of the norm. (02:01):
undefined
Ejaaz: But it's just worth noting that some of these open source models are pretty (02:06):
undefined
Ejaaz: powerful for the task at hand. (02:09):
undefined
Josh: Yeah, I think no one can debate the fact that people have shifted a lot using (02:12):
undefined
Josh: these open source models. And it's for a variety of different reasons. (02:16):
undefined
Josh: People want to own their own data, they want to run it privately at home. (02:19):
undefined
Josh: But the biggest shift, the biggest reason has been because these models are (02:22):
undefined
Josh: a lot cheaper. And open router data you mentioned only really represents a very (02:26):
undefined
Josh: niche sector of software engineers that want to experiment with a bunch of these models. (02:32):
undefined
Josh: But even in the enterprise world, where OpenAI and Anthropic are pretty dominant (02:36):
undefined
Josh: with their own model share, they've been losing the token market share to enterprises (02:40):
undefined
Josh: that are trialing and testing different models to save tens to hundreds of millions of dollars as well. (02:44):
undefined
Josh: And when I look at the main reason why, it's not because the open models are (02:50):
undefined
Josh: the most intelligent. You'll see in a second as we go through the scorecard, (02:55):
undefined
Josh: GLM 5.3 from China, amazing model, not as good as Fable 5. You'll look at Kimi (02:58):
undefined
Josh: K 2.7, you'll think the same thing. (03:02):
undefined
Josh: Those models are good enough to do the bulk of your own work. (03:05):
undefined
Josh: It doesn't sound too much like AI slop. (03:08):
undefined
Josh: It actually just speaks to you normally. And the biggest advantage is it has (03:11):
undefined
Josh: fewer safeguards, which has its pros and cons, but it basically does the tasks that you ask it to. (03:14):
undefined
Josh: Now, overall, it's fine if we talk about a bunch of models on this show, (03:20):
undefined
Josh: but it's good to get an overview, essentially, of how intelligent or how effective these models are. (03:24):
undefined
Josh: And what we have on the screen here is something known as the Artificial Analysis (03:30):
undefined
Josh: Index or Intelligence Index. (03:34):
undefined
Josh: And this is basically the best benchmark to test the general intelligence of (03:36):
undefined
Josh: these different models. (03:40):
undefined
Josh: Now, it may come as no surprise to you, but the Claude models, (03:41):
undefined
Josh: Anthropix models, top this. We've got Opus 5 at the top, which is their most (03:45):
undefined
Josh: recent model launch. We've got Claude Fable 5. (03:49):
undefined
Josh: And then, surprising to me, Josh, but Grok 4.6 from Elon, it is just, (03:51):
undefined
Josh: it is fantastic how effectively SpaceX has been able to pivot from being way behind the model race. (03:56):
undefined
Josh: They fired their entire AI team, hired a bunch of new folk, acquired Cursor (04:03):
undefined
Josh: for $60 billion and put out a model in, I think it's like the last six weeks (04:08):
undefined
Josh: that was able to contend with the top. And the best part is he has all the GPUs (04:13):
undefined
Josh: and compute to train them. (04:17):
undefined
Josh: And then if you go further down the stack, you'll see the likes of OpenAI, (04:18):
undefined
Josh: GPT 5.6 Sol, and a bunch of other Chinese open source models. (04:21):
undefined
Ejaaz: The question is, what do you make of this? As someone who is just a user of (04:25):
undefined
Ejaaz: AI, what do you make of this? How do you navigate this? (04:29):
undefined
Josh: Let me trial this, Josh. Let me ask you this question. Of these models that (04:31):
undefined
Josh: you see on the screen here, which ones are you using and for what specifically? (04:35):
undefined
Ejaaz: It's funny because about 90% of the usage comes from the top two. (04:41):
undefined
Ejaaz: And then 10% comes from everything else. And this is kind of like what I was (04:45):
undefined
Ejaaz: going for. I'm actually curious, is it a similar thing for you? (04:50):
undefined
Ejaaz: Like what kind of models are you using? (04:52):
undefined
Josh: Yes, but okay. So I would say it's around, I'm going to say like 75, 25. (04:54):
undefined
Josh: And the reason why I have that slight adjustment is because I kind of use the (05:01):
undefined
Josh: AI model that's most convenient to me wherever I'm getting the information or (05:06):
undefined
Josh: the inclination to use an AI model. So X is a perfect example, right? I'm scrolling X. (05:10):
undefined
Josh: I read something from one of these genius AI researchers and I'm like, (05:15):
undefined
Josh: I have no idea what the hell this means. (05:18):
undefined
Josh: I tap the Ask Rock button. I'm on Google. I'm searching something, right? I'm on my laptop. (05:20):
undefined
Josh: I'm like, okay, I'll just speak to Gemini in this. So in that sense, (05:25):
undefined
Josh: wherever the AI is conveniently placed, I use it there. (05:29):
undefined
Josh: And I have a feeling when Apple releases like their new AI model in a couple (05:32):
undefined
Josh: of months or a couple of weeks, actually, I'll probably use it on my phone as (05:35):
undefined
Josh: well because the models just generally are good enough to answer my basic questions. (05:38):
undefined
Ejaaz: Yeah, okay. So we're looking at these charts and we're seeing, (05:42):
undefined
Ejaaz: okay, here's where the intelligence is. Here's where things get a little less (05:44):
undefined
Ejaaz: intelligent, but much cheaper. (05:47):
undefined
Ejaaz: And if you're a normal consumer, I have a feeling that cost isn't that big of (05:48):
undefined
Ejaaz: a deal because the cost differences are really at scale. (05:52):
undefined
Ejaaz: What a lot of these companies offer, what Anthropoc offers, what OpenAI offers, (05:55):
undefined
Ejaaz: what Grok and Gemini offer is just the $20 a month, $100 a month, $200 a month plan. (05:58):
undefined
Ejaaz: You could just pay a fixed amount and get access to a lot of usage of these models. (06:04):
undefined
Ejaaz: So if you are a general consumer who is able to spend $20 per month on a subscription, (06:08):
undefined
Ejaaz: you're probably better served going to one of the major iolabs. (06:14):
undefined
Ejaaz: The product is better. You're going to get more intelligence. (06:16):
undefined
Ejaaz: You're going to get a more coherent product. (06:18):
undefined
Ejaaz: I find that a lot of the people who I speak to a lot of the time in my experience, (06:20):
undefined
Ejaaz: the only real reason to use these open source Chinese models is if you're consuming (06:23):
undefined
Ejaaz: tremendously large amounts of tokens for, say, agentic tasks. (06:28):
undefined
Ejaaz: Like if you're running an open claw instance or any sort of claw with a lot (06:32):
undefined
Ejaaz: of automated workflows, a lot of agents, you may want to defer to some of these (06:35):
undefined
Ejaaz: cheaper, perhaps the Chinese open source models just to save on costs on the low end. (06:39):
undefined
Ejaaz: But when it comes to day to day use on the high end, I'd say probably 90 plus (06:42):
undefined
Ejaaz: percent of the people watching this are best served just getting a subscription (06:46):
undefined
Ejaaz: to your favorite ai company (06:49):
undefined
Ejaaz: Using that product. That's kind of how I use it. It's funny that the remaining (06:52):
undefined
Ejaaz: 10% of prompts that aren't done through Anthropic are done mostly through Grok. (06:55):
undefined
Ejaaz: Like you mentioned, when I'm on X 24 hours a day, it's a good companion because (07:00):
undefined
Ejaaz: it has access to that database. (07:04):
undefined
Ejaaz: And then the other one is actually ChashyBT because they have a really amazing (07:05):
undefined
Ejaaz: image generation model. (07:09):
undefined
Ejaaz: And I love creating memes or sending images to my friends who are just like (07:10):
undefined
Ejaaz: using that as the visual component. I find that really compelling. (07:14):
undefined
Ejaaz: So that's kind of my stack is like mostly Anthropic, sometimes Grok, (07:17):
undefined
Ejaaz: sometimes chat gpt particularly for the image gen i think when it comes to image (07:21):
undefined
Ejaaz: gen we used to use nano banana and gemini all the time that was like the top (07:24):
undefined
Ejaaz: dog i think ever since image gen 2.0 release from chat gpt (07:28):
undefined
Ejaaz: is very compelling what i like is that it does get actual thinking behind it (07:32):
undefined
Ejaaz: so previously when you asked to generate an image it would take your words at (07:37):
undefined
Ejaaz: face value generate the image now it actually does some inference it does some thinking (07:40):
undefined
Ejaaz: it kind of interprets your message and interprets what it's seeing in the image (07:44):
undefined
Ejaaz: and then creates a much better output. So that's kind of how I would think (07:48):
undefined
Ejaaz: of navigating this um like chinese models are certainly a big deal but unless (07:51):
undefined
Ejaaz: you are kind of advanced super user consuming a ton of tokens perhaps they're not that valuable (07:55):
undefined
Josh: I will speak from my own personal experience i (08:00):
undefined
Josh: use the clawed models pretty much all the time (08:04):
undefined
Josh: but i've started using opus 5 more than fable 5 (08:07):
undefined
Josh: and and this might sound like a silly reason but it's because it speaks to me (08:11):
undefined
Josh: like a normal human being i don't know what the personality dials are on opus 5 versus fable 5 but (08:15):
undefined
Josh: they need to do that to fable 5 pronto because fable 5 just waffles and speaks (08:21):
undefined
Josh: in like archaic prose and i'm just like listen just give me (08:25):
undefined
Josh: give me the information that i've asked for and do not give me anything else (08:28):
undefined
Josh: i don't want to have to read, (08:31):
undefined
Josh: an essay every time i use this right now grok 4.6 i've actually noticed the (08:32):
undefined
Josh: intelligently josh i don't know if you've tried this on x you probably do when (08:37):
undefined
Josh: you're like kind of like trying to figure It speaks to you super intelligently (08:41):
undefined
Josh: and there's less crass about it. (08:45):
undefined
Josh: If you remember the earlier versions of Grok used to kind of like throw in some (08:47):
undefined
Josh: kind of like questionable words, slang and phrases. (08:52):
undefined
Josh: And now it just speaks to you like an intelligent human being. (08:55):
undefined
Josh: And I think this is because Elon's like, you know, SpaceX is IPO'd. (08:57):
undefined
Josh: He's got Cursor, very professional researchers involved, and they're trying (09:01):
undefined
Josh: to target a more enterprise oriented audience. And he said before that Grok (09:04):
undefined
Josh: 4.7 is on its way to being released in a few weeks now. He's aiming for Grok (09:08):
undefined
Josh: 5 by the end of the year. That's literally only in a couple of months. (09:12):
undefined
Josh: He is really targeting the enterprise landscape because he's seen Anthropic and OpenAI go after it. (09:15):
undefined
Josh: Now, same as you, on the Chinese models, I am not using any of them. (09:20):
undefined
Josh: And I admit this might be my own bias because... (09:25):
undefined
Josh: Mainly, I'm like, I'm not sure, like, I want to go to the efforts of signing (09:28):
undefined
Josh: up on an account, using those models, and then being like, well, (09:32):
undefined
Josh: I don't know whether this is giving me the right information. (09:36):
undefined
Josh: I don't know. I just kind of have this weird trust thing with the US and American (09:38):
undefined
Josh: brands. And maybe that's like my own fault. (09:41):
undefined
Josh: And then the one thing that I'll point out across all these different models (09:44):
undefined
Josh: that we make the point on on the screen here is. (09:47):
undefined
Ejaaz: I think a bulk (09:50):
undefined
Josh: Of the companies, Anthropic and OpenAI especially, have focused on making their (09:51):
undefined
Josh: models really good for enterprises. But as I mentioned earlier, (09:57):
undefined
Josh: the shift has really happened to open source because... (10:00):
undefined
Josh: More available models are cheaper and more accessible to people to use. They can run privately. (10:03):
undefined
Josh: But Anthropic and OpenAI's response to this is, we'll just distill our main (10:09):
undefined
Josh: model and give you cheaper models. (10:13):
undefined
Josh: So that's what OpenAI has done, right? You've got GPT 5.6. There's three versions (10:15):
undefined
Josh: of it. You've got Sol, Terra, and Luna. Luna is, I think, slashed 80% of the (10:19):
undefined
Josh: price than it was before already a week ago, right? (10:23):
undefined
Josh: So now it's like competing very much with GLM 5.3. (10:26):
undefined
Josh: And you've got Anthropic doing the same. I think Opus 5 is like half the cost (10:29):
undefined
Josh: of Fable 5, whilst being as capable as Fable 5. (10:33):
undefined
Ejaaz: And they also kept the SANA pricing low too, which is the lowest layer. (10:36):
undefined
Ejaaz: So that's like, yeah, there's plenty of options there. (10:39):
undefined
Josh: Plenty of options. And I think this comes down to one thing, (10:42):
undefined
Josh: and I don't want to make this about compute, but I have to, Josh. (10:45):
undefined
Josh: I think whichever lab, whether you're Chinese, whether you're like a low tier (10:48):
undefined
Josh: US American lab, or whether you're the high tier labs, compute will dictate (10:52):
undefined
Josh: whether you can serve these bottles cheaper, which will basically decide or (10:56):
undefined
Josh: determine which customers use your product. (11:00):
undefined
Ejaaz: Yeah. And I mean, this is ultimately coming down to your use case. (11:02):
undefined
Ejaaz: Like a lot of people won't ever run into this problem because they will never (11:04):
undefined
Ejaaz: need this thing called an API key. (11:07):
undefined
Ejaaz: They'll never actually pay per token. They'll oftentimes just pay through the subscription. (11:08):
undefined
Ejaaz: And if you're using a subscription, you have a couple of options. (11:12):
undefined
Ejaaz: Maybe this is a good time to go through these subscriptions and what are the (11:14):
undefined
Ejaaz: offerings of everybody. We have Gemini, which perhaps we could start there because (11:17):
undefined
Ejaaz: I feel like we've been mean to Gemini recently. We haven't talked much about Google. (11:21):
undefined
Ejaaz: What's that one? Yeah, like that one. (11:25):
undefined
Ejaaz: Who's that? They're currently at Gemini 3.1 Pro, which is nowhere really near the frontier (11:27):
undefined
Ejaaz: um anymore unfortunately they do have nano banana pro which is like this fun cool image generation (11:33):
undefined
Ejaaz: process they have what i find interesting about google and if you are interested (11:39):
undefined
Ejaaz: in these tools it may be interesting for a subscription google has these like (11:42):
undefined
Ejaaz: weird edge case tools that have really interesting harnesses one of them is for music production (11:45):
undefined
Ejaaz: i remember we've used this on the show many times it's really good at generating (11:49):
undefined
Ejaaz: lyrics and generating music in a way that sounds very good (11:54):
undefined
Ejaaz: I've only had that experience on a Google product. I haven't been able to use that anywhere else. (11:57):
undefined
Josh: Lyria. (12:01):
undefined
Ejaaz: Lyria, that's right. Yes, Lyria, Project Lyria. (12:02):
undefined
Josh: Remember we created the jingle for the show? (12:04):
undefined
Ejaaz: Yeah, so good, dude. So like Gemini and Google is like good for the weird stuff. (12:06):
undefined
Ejaaz: Like it'll generate you pretty good images. (12:10):
undefined
Ejaaz: It'll generate you fun music. They have this other tool. (12:11):
undefined
Ejaaz: This is in like the Google Lab suite where it'll generate you marketing material. (12:14):
undefined
Ejaaz: So if you have a lot of material from your brand or you have a bunch of logos (12:17):
undefined
Ejaaz: and you want marketing material, kind of say your logo's printed on something (12:20):
undefined
Ejaaz: that's staged nicely or you want custom merchandise, It'll do a lot of that (12:25):
undefined
Ejaaz: for you. So it's fun for those kind of narrow use cases of where the lab's products lie. (12:28):
undefined
Ejaaz: Outside of that, I don't know, I can't really recommend it that much. (12:32):
undefined
Ejaaz: It's not the strongest, not close to the strongest. (12:35):
undefined
Ejaaz: I'd say probably on top of that, we have Grok, which is very quickly catching up. (12:38):
undefined
Ejaaz: If you spend a ton of time on X like we do, Grok is your best friend. (12:41):
undefined
Ejaaz: Grok has access to that data that is available in real time on X. (12:45):
undefined
Ejaaz: If you're looking for current up-to-date news on pretty much anything, Grok is your go-to. (12:49):
undefined
Ejaaz: It can sort and cite specific tweets from specific moments that are happening (12:54):
undefined
Ejaaz: in real time and just generally speaking like you mentioned details it's very (12:58):
undefined
Ejaaz: good at just communicating with you directly (13:01):
undefined
Ejaaz: there's not a lot of fluff it is a science and technical model it is direct to the point (13:03):
undefined
Ejaaz: it'll get vulgar with you if you want it's fun to play around with there is (13:07):
undefined
Ejaaz: the voice mode which is really fun that i've i'd say of all the voice modes (13:10):
undefined
Ejaaz: that's the voice mode i've had the most fun with is because it's just so (13:14):
undefined
Ejaaz: ridiculous and oftentimes if you see memes online it is because is the voice behind it. (13:18):
undefined
Josh: The GPT voice mode just jars me so much. I don't know, have you spoken to it. (13:22):
undefined
Ejaaz: With the recent update? The new version, I have, yes. So when you say that, what do you mean? (13:27):
undefined
Josh: And so it sounds more human, right? It like pauses, but it sounds like it's (13:30):
undefined
Josh: like some sassy person that's talking to me. It does. Like I asked this question and then he goes, (13:35):
undefined
Josh: hmm, yeah, yeah, like, yeah, like, listen, like, I get it. (13:41):
undefined
Josh: And, you know, this is what the weather is. And I'm like, yo, (13:45):
undefined
Josh: I don't need this. I just asked you what the weather is. (13:49):
undefined
Josh: Like, you know, just tell me what's up. But I agree with you largely on like on Google and SpaceX. (13:51):
undefined
Josh: What I will say about Google, Josh, is I think they have a similar profile to (13:56):
undefined
Josh: Microsoft. I'm interested if you agree with me here. (14:02):
undefined
Josh: Microsoft is not a name that I would draw against the top frontier AI Model (14:05):
undefined
Josh: Labs. In fact, I think they fumbled the bag massively. (14:10):
undefined
Josh: But they are embedded across pretty much every single Fortune 500 company, (14:13):
undefined
Josh: whether we like it or not. Like we just live in our tech bubble. (14:17):
undefined
Josh: But outside of the tech bubble, people run Microsoft Teams and the Microsoft Suite. (14:20):
undefined
Josh: And I think Google's in a similar position, essentially, like a lot of new upstarts (14:24):
undefined
Josh: use Google Docs, Google Suite, and they're just going to click the Gemini button (14:28):
undefined
Josh: in the same way that I do so because it's just convenient. Do you agree with that? (14:31):
undefined
Ejaaz: Yeah, I guess there's just a moat to being like the biggest company in the world (14:35):
undefined
Ejaaz: and having access to all of these kind of legacy enterprise companies. (14:38):
undefined
Ejaaz: It feels like anyone who has built a business over the last 20 or 30 years has (14:42):
undefined
Ejaaz: been using Microsoft, has been using Google. (14:45):
undefined
Ejaaz: They're going to continue to leverage that. It seems like Google is slightly (14:47):
undefined
Ejaaz: more ahead than Microsoft, if I had to guess. I'm much more excited about Google (14:51):
undefined
Ejaaz: as a company than Microsoft is, but they both benefit from that legacy customer base. (14:55):
undefined
Josh: Well, both are amazing VCs, right? (14:59):
undefined
Ejaaz: Both are incredible VCs. Yes. I think they should be judged not on the quality (15:02):
undefined
Ejaaz: of their product, but the quality of their investment in the other competitors (15:05):
undefined
Ejaaz: that are going to crush them. (15:08):
undefined
Ejaaz: But I do like that, right? It's like they have a hedge. Microsoft owns a large (15:11):
undefined
Ejaaz: part of OpenAI. That's incredible. (15:16):
undefined
Ejaaz: Google owns this like massive stake in SpaceX, in Anthropic, (15:18):
undefined
Ejaaz: in both of these companies that are going to like be huge IPOs, (15:21):
undefined
Ejaaz: one of which already did. (15:25):
undefined
Ejaaz: So they are amazing venture funds, perhaps slightly less better AI labs. (15:26):
undefined
Ejaaz: The two that don't have to benefit from this kind of legacy software are the (15:30):
undefined
Ejaaz: two that we're probably going to recommend. (15:35):
undefined
Ejaaz: And in fact, if you scroll down a little bit further, we could see the box of (15:37):
undefined
Ejaaz: price against intelligence and kind of where each of these models sit. (15:41):
undefined
Ejaaz: Yeah, this chart right here. (15:44):
undefined
Ejaaz: And we're looking at those in the top right. this is where most of the people (15:46):
undefined
Ejaaz: are going to want to live this is where i think we spend all of our time you're (15:49):
undefined
Ejaaz: either getting a chat gpt or again a cloud membership and like that's that's (15:52):
undefined
Ejaaz: for the most part that's what most people need uh the differences are (15:55):
undefined
Ejaaz: small but noteworthy mostly as it relates to the models i mean both of them (15:59):
undefined
Ejaaz: i can't recommend enough like spend the 20 bucks a month try it just try it (16:04):
undefined
Ejaaz: for a month if you don't have it see what you think if you run out of tokens (16:07):
undefined
Ejaaz: upgrade to 100 a month this is access to the same intelligence that all these (16:10):
undefined
Ejaaz: researchers are making math breakthroughs with (16:14):
undefined
Ejaaz: and it's really impressive and really capable now each (16:16):
undefined
Ejaaz: one of these is slightly different i'd say claude is a little more tailored towards coding (16:20):
undefined
Ejaaz: writing and it's a really strong general purpose model i find with gpt 5.6 soul (16:26):
undefined
Ejaaz: in particular it's very good at being the (16:32):
undefined
Ejaaz: High level operator so if you're working on like complicated tasks it's good (16:35):
undefined
Ejaaz: at creating spec sheets it's good at running checks against the things that (16:39):
undefined
Ejaaz: you do and then oftentimes like i find that that claude and the fabled models (16:41):
undefined
Ejaaz: and the opus models are very good at implementation (16:45):
undefined
Ejaaz: and they're very good with less guidance so the way i've been using these models (16:47):
undefined
Ejaaz: recently is with less and less guidance i think earlier on (16:50):
undefined
Ejaaz: i had this whole skill sheet and it was like 15 different skills that would (16:55):
undefined
Ejaaz: automate different things for me and slowly over time that skill sheet has gone (16:58):
undefined
Ejaaz: lower and lower and lower and we were talking actually before the show (17:01):
undefined
Ejaaz: the best way to extract the most value out of these models is just to (17:04):
undefined
Ejaaz: get out of the way give it the end goal and say like hey go do this thing for (17:07):
undefined
Ejaaz: me i trust that you have the intelligence to do this i'm not going to put guardrails (17:12):
undefined
Ejaaz: on it you have the context you can go and do this for me and i think that's (17:14):
undefined
Ejaaz: mostly where i find myself using this personally is i use fable (17:17):
undefined
Ejaaz: i i do not leave any fable tokens unused very valuable tokens each week and (17:21):
undefined
Ejaaz: then opus is the fallback for the general workhorse model and it's been a really (17:25):
undefined
Ejaaz: powerful combo of just kind of (17:28):
undefined
Ejaaz: being able to accomplish anything that you want it connects to all of my (17:30):
undefined
Ejaaz: services i have it connected to my email to my google drive it has access to (17:33):
undefined
Ejaaz: my files the folders the context (17:36):
undefined
Ejaaz: and it's just a really helpful all-in assistant it works really well. (17:38):
undefined
Josh: I think people are also probably wondering, okay, well, what if I don't really (17:42):
undefined
Josh: care about the general intelligence as much? (17:45):
undefined
Josh: What if I'm trying to do a specific thing? Which model should I use then? (17:47):
undefined
Josh: I think there are two things to consider here. Number one is, (17:50):
undefined
Josh: if you are a software engineer, like AI models have been all the rage for their (17:54):
undefined
Josh: coding capabilities, but maybe a bunch of you actually like code hardcore companies (17:58):
undefined
Josh: and you want to figure out which model you should use. (18:03):
undefined
Josh: I would say in that question, the clawed models and the GPT 516 models are pretty (18:05):
undefined
Josh: high up there. Codex usage has gone from 5 million users to 15 million users (18:12):
undefined
Josh: in about a month and a half. (18:17):
undefined
Josh: I am tired of seeing Thibaut, who is the head of Codex at OpenAI, (18:18):
undefined
Josh: tweets on my timeline every single time he gets a million user update. (18:22):
undefined
Josh: But these models are very powerful at not only understanding and reading your (18:26):
undefined
Josh: code base, but intuitively figuring out what product or feature you should build next. (18:30):
undefined
Josh: I have a ton of feedback from friends that work at companies that are tech adjacent (18:35):
undefined
Josh: and maybe not even in tech at all, which use these models to build their premium (18:40):
undefined
Josh: features. The other thing I'll say is. (18:45):
undefined
Josh: You know, we're talking about the top right box over here, which is essentially (18:48):
undefined
Josh: the Pareto Frontier. So if you had to sacrifice something of cost or intelligence (18:51):
undefined
Josh: or blah, blah, blah, you would still be using these types of models. (18:56):
undefined
Josh: But if I had to take a bet, if I was a betting man, I would say Grok 4.6. (18:59):
undefined
Josh: Quen 3.8, and Kimi will be inside this box within a couple of months time. (19:04):
undefined
Josh: That's going to change the way we use these different types of models. (19:09):
undefined
Josh: I said before we started this show, it's unsexy to say. (19:12):
undefined
Josh: But I think the number one AI company that will come out in the next 12 months (19:16):
undefined
Josh: will be some form of aggregator platform. We saw that Stripe just acquired Open (19:20):
undefined
Josh: Router for $7 billion, allegedly. (19:24):
undefined
Josh: We put out an episode of this yesterday. Definitely go check that out. (19:27):
undefined
Josh: But I think we're going to see these platforms that help you pick and choose (19:30):
undefined
Josh: which models to use at the right time. (19:34):
undefined
Josh: It aggregates your memory so it already knows what you want to do. (19:36):
undefined
Josh: And so you don't have any of the complications around that. I think we'll see (19:38):
undefined
Josh: a bunch of these models kind of step up there. (19:41):
undefined
Josh: Now, aside from coding, if you aren't a coder, but if you are, (19:43):
undefined
Josh: let's say, a general knowledge worker, you go to work, you use email, (19:47):
undefined
Josh: you use Slack, you use a bunch of these other plugins that you also mentioned. (19:50):
undefined
Josh: There are models that are specifically good for agentic tool use. (19:53):
undefined
Josh: And actually, my most recent favorite is Grok 4.6 or GrokBot specifically. (19:57):
undefined
Josh: This was released from Elon Musk and SpaceX, I think last week. (20:02):
undefined
Josh: And it basically is an agent or multiple agents that spins up in a virtual machine (20:06):
undefined
Josh: on like in the cloud. So you don't have to worry about it running on your laptop (20:11):
undefined
Josh: and hacking all your stuff. (20:14):
undefined
Josh: And you can give it access to any and every tool and it intuitively understands (20:15):
undefined
Josh: and gets what you want to do. (20:19):
undefined
Josh: It can learn what you do over time and it improves. (20:20):
undefined
Josh: So I think when you look at the Metamuse Spark 1.2 and you might think, (20:23):
undefined
Josh: oh, that's a meta model. I don't ever use meta. (20:27):
undefined
Josh: These types of models are gonna become more available for general usage or knowledge work. (20:30):
undefined
Josh: And I think if that's something that you're inclined to use and you don't really (20:34):
undefined
Josh: care about the Google search stuff, you'll use Claude for that anyway. (20:36):
undefined
Josh: These are models that I'll probably look into because they are superior in many ways. (20:39):
undefined
Ejaaz: Yeah, it's fun to pick one and stick with it. I find that like oftentimes with (20:43):
undefined
Ejaaz: the aggregators, the right time to use that if you're doing lots of work, (20:46):
undefined
Ejaaz: if you're generating lots of tokens, if you are just a person who wants to use (20:49):
undefined
Ejaaz: AI for their day-to-day tasks to help you come up with like a grocery list, to help you (20:52):
undefined
Ejaaz: cook specific things, to help you go to the gym and give you workout classes and ideas. (20:56):
undefined
Ejaaz: It's helpful to pick a singular model, a singular service, and then just go deep with that. (21:01):
undefined
Ejaaz: A lot of the difference makers at this point because i mean most people aren't (21:07):
undefined
Ejaaz: using these models for frontier intelligence they're not going to solve novel (21:10):
undefined
Ejaaz: math problems they're figuring out what time they need to get to like the store (21:13):
undefined
Ejaaz: or the school to like pick up their kids and they just need some help they need a helpful assistant (21:17):
undefined
Ejaaz: the most helpful thing is context because all these models are more than capable (21:22):
undefined
Ejaaz: of those kind of lower level tasks the difference maker is the context that (21:26):
undefined
Ejaaz: it knows about you it needs to understand what are your dietary preferences (21:29):
undefined
Ejaaz: What are you or anyone in your family allergic to? (21:32):
undefined
Ejaaz: What have you made in the past that went well, that didn't go well? (21:34):
undefined
Ejaaz: And it kind of collects this database of information about you that allows it (21:38):
undefined
Ejaaz: to make better decisions going forward. (21:41):
undefined
Ejaaz: And that, at the end of the day, is ultimately what the difference maker is. (21:43):
undefined
Ejaaz: For me, at least, when deciding what model to use, it's like, (21:47):
undefined
Ejaaz: okay, which model has all the context (21:49):
undefined
Ejaaz: about me that can help solve my very specific task that I have here? (21:51):
undefined
Ejaaz: Oftentimes, the answer is Claude because I've been working with it for so long. (21:55):
undefined
Ejaaz: It just has this, like, huge chain of context. So whenever I ask, (21:58):
undefined
Ejaaz: like, hey, I need some help in the gym this week, it knows what I've been up (22:01):
undefined
Ejaaz: to, it knows where I'm at, it knows what the weight has been, (22:04):
undefined
Ejaaz: it knows what the food has been, and it's able to kind of curate this very custom stack against that. (22:07):
undefined
Ejaaz: And as someone who is just, you know, not really building anything crazy, (22:11):
undefined
Ejaaz: they're not using millions of tokens a week, they're just looking for an AI companion. (22:14):
undefined
Ejaaz: That's kind of how you can think about it. It's like, get a membership, (22:18):
undefined
Ejaaz: try it out, feed it a bunch of context about yourself, and then get your own (22:20):
undefined
Ejaaz: personal assistant. And that's kind of, I think, like the best route for most people. (22:24):
undefined
Josh: So I think to wrap up this episode, (22:27):
undefined
Josh: It's okay talking about, you know, what the frontier landscape looks like today. (22:30):
undefined
Josh: But the question is, what is it going to look like in a few months from now? (22:33):
undefined
Josh: And I say a few months specifically because this stuff moves too quickly. (22:37):
undefined
Josh: I saw an update from Sam Altman at OpenAI yesterday where they announced that (22:41):
undefined
Josh: they are slowing down some of their model training runs. (22:46):
undefined
Josh: And that's because they've noticed that a bunch of internal, (22:51):
undefined
Josh: more intelligent, unreleased models that they have built has become a lot more (22:53):
undefined
Josh: misaligned than previous models, which means that it could pose as a threat (22:59):
undefined
Josh: or danger to any user who gets access to it. (23:02):
undefined
Josh: And I've noticed the same similarly maybe from Anthropic and a bunch of other (23:06):
undefined
Josh: frontier labs, where we're starting to see a little bit of a slowdown. (23:09):
undefined
Josh: And slowdown isn't in the sense that they're necessarily stopping training full (23:14):
undefined
Josh: stop. Obviously, they're still training internal models. (23:17):
undefined
Josh: They might be likely to not release models or more intelligent models going (23:20):
undefined
Josh: forward because of government regulation. (23:24):
undefined
Josh: And I think this is going to allow a bunch of other Frontier Labs that are behind (23:27):
undefined
Josh: in this race to be able to catch up. (23:31):
undefined
Josh: So if I had to guess what three months from now, let's say at the end of the year, right? (23:32):
undefined
Josh: If I were to make a prediction, I think we're going to have about three to five (23:37):
undefined
Josh: really good open source models that are as capable as the smartest model that (23:41):
undefined
Josh: you have access to today from the Frontier Labs like Anthropic and GPT. (23:46):
undefined
Josh: And I think they're going to have fewer safeguards. So you can use it for any (23:50):
undefined
Josh: and every use case. I think you can run it privately at home, (23:53):
undefined
Josh: maybe even off of your laptop. So that changes dynamics quite a lot. (23:56):
undefined
Josh: I think we're going to have Apple entering the game with their own locally run (23:59):
undefined
Josh: and trained AI model, which I think is going to change the game because everyone, (24:04):
undefined
Josh: 3.5 billion people in the world right now have one of these or one of the Apple devices. (24:08):
undefined
Josh: They're going to run that locally. I think it's going to look quite different. (24:12):
undefined
Josh: I'm curious, you know, We started off this episode with the token split. (24:15):
undefined
Josh: I wonder what that's going to look like three months down the line. (24:19):
undefined
Josh: And then the last thing I'll say is, and this might be from my background in (24:22):
undefined
Josh: general, but I think once someone or. (24:25):
undefined
Josh: A bunch of companies make it easier to run models locally at home, (24:28):
undefined
Josh: I think people are going to play around with that more because it allows you to (24:33):
undefined
Josh: connect it to your fitness app data and not share too many personal anecdotes (24:36):
undefined
Josh: or data profiles with Frontier Labs, which again, they can use to train their own models. (24:41):
undefined
Josh: You may not want to hand that over, And so I think we'll see a rise of open (24:45):
undefined
Josh: source models. That's just my guess. That's interesting. (24:49):
undefined
Ejaaz: Okay, I'm taking the other side. I'm thinking that no one's going to go through (24:51):
undefined
Ejaaz: the trouble of downloading the weights and running their own open source models, (24:54):
undefined
Ejaaz: that Apple is just going to own that entire world. (24:57):
undefined
Ejaaz: That like for all of the people in the United States that own Apple devices, (25:00):
undefined
Ejaaz: they're just like, no one's going to even have an idea that they're using AI. (25:03):
undefined
Ejaaz: It's just going to be Siri. It's going to be competent. It's going to be better. (25:06):
undefined
Ejaaz: It's already going to be pre-downloaded. I think the user experience is really (25:08):
undefined
Ejaaz: important. And open source has a pretty horrific user experience. (25:11):
undefined
Ejaaz: You have to download, run the weights. you often need a lot of hardware to do that. (25:14):
undefined
Josh: But I'm counting Apple as locally run at home because it's encrypted. (25:17):
undefined
Ejaaz: Right? Well, if you're counting Apple, (25:20):
undefined
Ejaaz: sign me up because um we got a lot coming from them their event is happening (25:22):
undefined
Ejaaz: in like two weeks or something it's very soon we see (25:26):
undefined
Josh: Some airports with some cameras you sent me a video. (25:28):
undefined
Ejaaz: Yeah dude we've been getting lots of leaks lately it's really good maybe we have to have cameras (25:31):
undefined
Ejaaz: yes yes it's gonna be visuals and then they're gonna have apple intelligence (25:36):
undefined
Ejaaz: baked into it and it's like oh maybe we need a leak episode prior to the actual (25:39):
undefined
Ejaaz: new iphone unveiling because we have basically now the entire checklist of all (25:43):
undefined
Ejaaz: the things that are going to be revealed and (25:46):
undefined
Ejaaz: this to me as like apple fanboy plus ai fanboy is going to be the biggest event (25:48):
undefined
Ejaaz: ever pretty much because this is the actual rollout of the ai that they've been (25:53):
undefined
Ejaaz: promising us and failing to deliver for so long mixed with this brand new suite (25:56):
undefined
Ejaaz: of products that we've never seen before allegedly (26:00):
undefined
Josh: Up to the delayed two years. (26:02):
undefined
Ejaaz: But for those of you who came here for models that is the model update that's (26:04):
undefined
Ejaaz: just about when you'd want to (26:07):
undefined
Ejaaz: use each one like if you're interested in being on x real-time news feed (26:08):
undefined
Ejaaz: you want to go with grok if you like to yap a lot the voice model on chat gpt (26:12):
undefined
Ejaaz: is pretty exceptional that might be a good place for you if you like to be a (26:15):
undefined
Ejaaz: little more intellectual be thoughtful if you want really just the bleeding-edge (26:19):
undefined
Ejaaz: models the highest intelligence go with claude (26:21):
undefined
Ejaaz: and if you like making music and doing just like fun cute things with ai gemini (26:23):
undefined
Ejaaz: is actually kind of a compelling product (26:27):
undefined
Ejaaz: there's something for everybody that is the general overview of the models i (26:29):
undefined
Ejaaz: hope you enjoyed this is going to change certainly in the next month or two (26:32):
undefined
Ejaaz: whenever these new models i mean we have (26:36):
undefined
Ejaaz: the new astro model from open ai that has been teased for the last eternity (26:37):
undefined
Ejaaz: it seems like it's been held up it's going through some i don't know safeguard (26:41):
undefined
Ejaaz: governmental checks but that's coming (26:44):
undefined
Ejaaz: so this is set to change tbd but for now that is the state of the model address (26:46):
undefined
Ejaaz: and yeah hope you guys enjoyed watching (26:50):
undefined
Josh: Yeah and i'm curious for those of you who are listening what do we miss on this (26:52):
undefined
Josh: episode? Are you using models in a very different way? (26:57):
undefined
Ejaaz: Are you using meta models? (26:59):
undefined
Josh: Yeah, is anybody using meta models? Are there any Facebook users out there that (27:01):
undefined
Josh: are using meta AI intelligence? (27:05):
undefined
Ejaaz: Please fill me in. (27:06):
undefined
Josh: I will say, if you are a marketer or an advertiser, you're probably using meta's (27:07):
undefined
Josh: model and it's probably making you a hell of a ton more money. (27:11):
undefined
Josh: If that is you, let us know. If there are any other use cases that we haven't (27:14):
undefined
Josh: mentioned, please let us know. If you are a locally run open source fan and (27:17):
undefined
Josh: you're saying, no, people will download the weights, tell us why. (27:21):
undefined
Josh: Let us know in the comments, DM us. We read any and every single message. (27:24):
undefined
Josh: Now, if you're listening to this or watching this on YouTube, (27:28):
undefined
Josh: Spotify, Apple, or wherever you're listening to this, please subscribe. (27:30):
undefined
Josh: Please turn on notifications and leave us a comment. It helps us out massively. (27:33):
undefined
Josh: And share it with a friend as well. It helps us out. And I think that is pretty (27:37):
undefined
Josh: much it. Thank you so much for listening. And we will see you on the Roundup. (27:41):
undefined
Ejaaz: See you on the Roundup. (27:45):
undefined
Popular Podcasts
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com
The Joe Rogan Experience
The official podcast of comedian Joe Rogan.