I.e. in the early days (2023-2025, lol), companies were mostly differentiated by the quality of their LLM, i.e. GPT-3.5 was leagues ahead of anyone else, so was for serious work the only game in town. Until recently it was basically only the US based companies (OpenAI/Anthropic) with models worth using.

Now, Anthropic and OpenAI are neck-and-neck, and Chinese models are catching up fast, and the models are becoming commodities, and as long as they reach some threshold of quality/capability, it doesn't really matter to most customers.

So now we might be reaching the branding stage of LLMs. I.e. do you go for Anthropic, who are branding themselves as the responsible ones, or OpenAI, who are branding themselves as... the innovators? Not sure what they are going for these days.

If all models are roughly equally capable, all that's left is branding.

At least those are the thoughts it provoked in me. If that's what Paul Graham meant or not I have no idea.