AI models are becoming capable enough to use existing software

Transcript

Transcribed by machine (Whisper) from the episode’s audio, so a word may be off here and there.

1. 1:04:23

   GPT 6.1 Terra, which was released yesterday. A quick reminder, OpenAI has four tiers. Astra is their flagship. Sol is their high-performance. Terra is their everyday workhorse. And Luna is their budget model. So two numbers jumped out, for me at least, on GPT 6.1 Terra. First is the price. It's come in at less than half the price of the previous Terra model. And then hallucinations. GPT 6.1 Terra demonstrated a 50-point drop in hallucination.

2. 1:04:55

   This is a demonetization curve. real-time, right? Every few months, the same intelligence costs half the price. Alex and Imad, your take on these models. Alex, you first. We're right now in a cycle, which I think a hyper cycle, mega cycle, if you will, that's coming to an end where both OpenAI and Anthropic have gotten into the habit of releasing larger models. And then in sort of a tick versus talk, if you will, cadence, they release the largest model first, and then they release presumably smaller distillations internally of the same model.

3. 1:05:31

   And GPT 6.1 Terra is, to my eye, a relatively unexciting distillation of earlier Astra-slash-Sol level or sized or footprinted models. Obviously, the smaller, more distilled variants that occupy a different position on the cost versus performance curve tend to be more exciting for embedded, high volume, low dollar value per token use cases. So I tend to get a lot less excited over these sort of late curve, late cycle distillations. The good news, the solace I draw is the fact that OpenAI has gotten around to releasing the Terra version of 6.1 means that maybe we'll see 6.2 or 7 sometime soon.

4. 1:06:20

   Do you mind anything exciting there for you? Yeah, so I think that with Terra and Sonnet 5, you're seeing the models that are being lined up to be the personal companion models. And those need to just be really solid. This is why the hallucination notes drop. Right now, if you use DOTS, which is OpenAI's companion, it uses Astra, so you actually have unlimited Astra via that. But that's not sustainable. Again, everyone's trying to drive down the price of that.

5. 1:06:45

   except for if you come to europe so i have dots in the us and then when i come to the uk it turns off so that was very interesting come to come to america imad we want you come to those shows um so again i think you'll see a bifurcation of models they'll be the model that's closest to you that's really reliable and they'll optimize for that and it'll pull in the smart models it'll pull in the models with artistic taste it'll pull in the models that are really good accountants and you won't really see much more again we're moving towards this control plane type of scenario two things uh that struck out for me one uh the day the days between gpt astra and terra were 53 days for a 50% drop in the token price. We've been looking at a 10-week doubling. This was a 7 and a 0.4-week doubling, right? So now we've dropped the doubling time for price performance

6. 1:07:40

   to seven weeks, pretty much. That's pretty insane. That was one. But I think more importantly, as the thresholds that these models are reaching, you're at a point where you don't need to redesign software for AI. It's going to operate the software it has. It's just going to start to exposing more and more how inefficient our organizational designs are. And so, we've been looking at that a hell of a lot. So, amazing at all levels. I can't wait to get my hands on it.

7. 1:08:05

   Yeah, I think it's an important point there. Yeah, important point that Ahmad and Salim made that, you know, for personal agent use, these things are over the bar of just brilliant, like just incredible. And so, this curve is almost at its end in terms of the personal home agent. And the next curve is really accelerating, which is its access to every kind of tool you can imagine. So all the post-training and all the MCP wrappers are now embedded, and it just happened kind of quietly. Now, if you pick up one of these models and you say, hey, can you access 11 Labs and make a really cool video with great voice overlay, and I want it to be my voice, it just happens.

8. 1:08:42

   Six months ago, you had to point it to an API, you had to coach it, you had to, and now it's just out-of-the-box capabilities. So I think the new curve will be a different dimension, which is, you know, consumer satisfaction, consumer happiness, consumer, all the tools already in the box, you know, all those features. It's very similar to when the PC, you know, PCs used to be so slow and crappy that they were marketed as 40 gigahertz, you know, 50 or 40 megahertz, 80 megahertz, 120 megahertz. And then Apple said, you know, nobody gives a rat's ass about megahertz anymore.

9. 1:09:14

   They care about iTunes. They care about whatever. They care about whatever and completely flipped the paradigm. I think we're right at the edge of that now with these models where they're all more than smart enough. And now it's like, what does it do for you out of the box? I think we'll see a new curve. This episode is brought to you by Blitzy,