20 chunks · 17 episodes
... layer was laid down by a human mind, nesting and abstracting. But systems like LLM models are like programs that were grown via process. The models are thus like a programming language that no human mind will ever be able to comprehend.
Is the LLM model above or below the API in your system? Is the agent swarm above or below the API? Does a user have to think about it?
...ders have to compete and none of them win" for LLMs. Great for everyone but the LLM model providers, who are in a never-ending red ocean battle. But the rest of us benefit from the significant competition.
The LLM model providers are like electricity providers back when electricity was new. Competing to get better quality for cheaper. Innovating on new techniques to ...
Stuart Russell has an evocative metaphor for the dangers of large LLM models. He compares them to flying passengers via a message bird. Imagine stuffing all of the humans into a capsule flown by a massive bird. That would be ...
Is the LLM model a Christmas tree decorated with doo dads? That is, like a Chatbot? A vertical experience. Or is it like electricity that can be infused into everythi...
I continue to think the best business parallel for LLM model providers is cell phone networks. Extremely capital intensive to build out, but then much lower marginal cost to operate. Though inference has much m...
...ness value now comes mostly from the consumer subscriptions, not the underlying LLM model. They started off by having the first break-out model quality. But now their value is less the model (there's a whole peloton of similar-quality comp...
...ex specialized stuff. But LLMs might not need that. It's kind of weird that the LLM model creators also have consumer frontends to them. It shows how powerful LLMs are that they can be a viable product even used directly by users.
Companies trying to differentiate an LLM model in their niche vertical risk general purpose models eclipsing them.[iy] Only the big kids can compete in the general model race, because it's so reso...
The equilibrium of the best LLM models being available via API seems meta-stable to me. You could imagine an alternate universe where ChatGPT got popular before OpenAI had released a publ...
LLM model quality seems to be reaching an asymptote. You can only see the difference between models after multiple conversation turns now. This is good for eve...
...ados can distinguish small quality differences. The same is true for coffee and LLM models. They have enough experience with the various options and calibrated taste to be able to distinguish subtle differences and have informed preference...
...o a competitor. This makes them more commoditized than they otherwise would be. LLM models don't store any state, are highly commoditized, and are also insanely capital intensive to set up. Not a great business!
...ayer from the application layer. The model layer is the creator and operator of LLM models. The application layer is the creator of the UX that actual end users use. These are two extremely different layers. They are different pace layers....
... looks quaint. You can't control a number, the asymmetry is too strong to leak. LLM model weights are just really really really big numbers.
... but it will take time to discover which ones. We are in the early innings! The LLM model providers are the electricity providers. Expensive, competitive, value-creating... but not necessarily a great business.
... that world! But less strategic power than the things that directly face users. LLM model providers will definitely be important... and also more likely to be subterranean. Unless the UX of actually using the models, e.g. high-quality inte...
Normal LLM models are like the brain's System 1. That is, highly parallel, vibes matching from past experience. OpenAI's o1 (codenamed Strawberry) is different: like ...
OpenAI and Anthropic had an underlying LLM model that was so good that they could slap on a demo level of UX and it was a viable product. But they are not differentiating on UX ability, that is not ...