I was skeptical about the results because in my experience both recent GPT and Opus modules are strong. Everything else is B or C tier. This is just artisanal vibe testing though. It's very hard to eval them properly.

I'm very interested in APIs that allow client-side context construction rather than relying on opaque APIs concatenating strings from your JSON messages and injecting tool prompts. I found that generally, you can craft the entire context as a unicode string and just stuff it in the system message. This works best with models where the chat template is published.

Overall I find their API design and docs so messy. It's a shame, since it's the main entrypoint to using their service.

As you mentioned, they know they need good data though, so they might actually try to find some equilibrium.

If not, it's possible that the creation of new valuable content, to feed the LLMs, will be produced in-house by the AI labs. Sounds insane, but Netflix also makes their own content.

I think the AI labs will become so big that they'll take on more roles than just offering LLM inference. I think they'll become as or more powerful than many current nation state governments.

I've used Win/Mac both very extensively as daily drivers, but since 2021 using exclusively Ubuntu LTS. In practice I don´t notice any lesser quality, or drivers not being good. I basically have no complaints. I'd say my Ubuntu 22 LTS is a lot less buggy than the megacorp stuff I used to use.