Simple solution to slow down recursive AI self-improvement: – The lab with the top-ranked model must accept that THEY should not use it to work on cutting-edge AI – But everyone else should have access to it. By definition, this
@jeremyphoward
-
Anthropic documented cyber warning and LLM improvement sabotage
By
–
Yes there's 2 separate pieces: 1) if apparently doing cyber sec stuff, downgrade to opus and warn; 2) if apparently trying to improve frontier LLMs, silently sabotage the work. (That 2nd one reads like an insane conspiracy theory! But it's actually documented by Anthropic.)
-
AI to become a rare diamond in mediocrity
By
–
Moreover, those who focus on using AI to help improve the skills of themselves and their teams will be highly sought-after diamonds, as they will be the rare A++ players in an ocean of mediocrity.
-
Using Kimi, MiMo, Deepseek but text-only limitation
By
–
We already do, but not via a gateway – we do it ourselves. Kimi and MiMo are both great. Deepseek is too, but it's text-only, which is a very significant limitation for us.
-
10x more completions than Opus in practice
By
–
Yes I know how it works, but in practice it's about 10x more completions than using Opus for the same task, in my experience. You really need to try it for yourself to see. It's night and day.
-
GPT-5.5 cheaper via API subscription, more efficient thinker
By
–
No. GPT-5.5 is *much* cheaper building against the API directly, when using a subscription. And it's somewhat cheaper even when not using a subscription, because the Opus 4.7/4.8 is much less efficient, and because GPT-5.5 is a more efficient "thinker".
-
Anthropic criticized for failing to lower API costs
By
–
Has @AnthropicAI completely given up on making API usage reasonably-priced? Following the token-usage changes recently they announced various updates to *subscription* usage to make it more reasonable. But they've done NOTHING for API users. The cost is insane at this point.
-
Jeremy Howard now using GPT-5.5, likes it nearly as much as Opus 4.6/4.7
By
–
I've largely switched over to using GPT-5.5 in recent weeks, which I like nearly as much as Opus 4.6 and 4.7, and is *very* reasonably priced (since I can use my subscription with the @OpenAI API.)
-
Model 4.7 preferred but GPT 5.5 better value for price
By
–
I still enjoyed 4.7. It remained my preferred model until today, although I liked GPT 5.5 nearly as much (and greatly preferred it when taking price into account, since it can be used with API at subscription pricing).
-
Opus 4.8 more cooperative but still expensive
By
–
Worked on some code this morning using Opus 4.8 and so far I'm really liking it. Much more cooperative than 4.7 and less "over agentic". Stops and asks for my input when needed in places 4.7 (and GPT 5.5) would just foolishly blast ahead. (Still WAY too expensive.)