People often keep bringing up the example of chess as proof that, even if in the initial states humans+AIs will be superior to AIs alone, as evidenced by chess this too shall come to pass, as AIs alone eventually surpassed human-assisted teams.

However the timelines for how fast this happened seems to have been compressed in everyone’s minds. Astra’s estimate is that this took 15-20 years, counting from the moment of Deep Blue’s victory over Kasparov.

You could, of course, argue that we now have a lot more more money and talent poured into AI, but at the same time chess is also a far simpler domain for computers to handle, so it’s quite possible that humans+AI “teams” will be better than “AI alone” for at least a decade. Whether or not such teams will be able to compete with AIs alone economically speaking is, of course, a different question, but in terms of ”who’s strictly better” the timelines might be longer than you expect.

If you have a Substack blog, consider using Datawrapper embeds for inserting tables or charts. For some reason Substack has been somewhat hiding this cool feature as its not visible in the menu and poorly advertised but it lets you create highly customizable charts/tables for your Substack posts - and is also supported on Wordpress or in the form of an HTML embed if you have a self-hosted blog/website. I wrote a brief guideline on how to use it in my blog - and you can see cool Datawrapper embeds in Nate Silver's substack.

Sadly Datawrapper not supported on LessWrong (yet) but LW already has native support for table embeds, so its less relevant.

I’m surprised that many people acting as if the HF-OpenAI incident presents a risk to humanity in the near future (1-3 years), rather than something that will only have consequences long down the line (10-15 years). Right now even the worst case scenarios don’t represent a risk of extinction, unless you count in the “AI makes a supervirus” risk, which seems far fetched to me as of 2026.

All HF-OpenAI did was demonstrate that the LW/EY predictions were basically correct, but I don’t think anyone’s p(doom) or timelines should be updated as a result, because, well, we’ve all pretty much expected this to happen, so a good Bayesian wouldn’t update much on expected evidence.