8 Comments

This post provides me a lots of insights. Whenever I see a thing that is better and cheaper, my first thought is whether it is too good to be true. This post helps me understand such a result better.

Expand full comment

I’m thinking the question what is the moat for the AI model companies. From the paper, it reads like the amount of accessible compute is the moat. Does it imply that the near term ai innovation is determined by the compute at large?

Expand full comment

Users and user data. ChatGPT.com is the moat -- no reason to go elsewhere for most people.

Expand full comment

Given the ai chip ban, China could enforce everyone in China to use deepseek or qwen, which also has access to a good user base. Basically, I want to answer a high level question: is the ai gap between US and China increasing or decreasing?

Expand full comment

Am working on a piece on this, at least in the open source, with Jordan Schneider of China Talk. I actually think the gap is... steady. China is closer than most people expected and not falling behind? Is that a cop out answer?

Expand full comment

Yes, this is a good answer. I am a bit paranoid on understanding the gap in a normalized setting. I don’t like exaggerated marketing announcements and don’t want to be fooled too much. Anyway, it is a personality.

Expand full comment

IMO being first-to-market for consumers and evolving platform features for lock-in will be the enduring moat for OpenAI, in a way that's not as appreciated right now because of the model arms race.

Expand full comment
Comment deleted
Jan 11Edited
Comment deleted
Expand full comment

all tracks... I'd think the best option to start with is to just try a detailed system prompt :)

Expand full comment