New Dario essay is pretty wild
V thought-provoking
https://x.com/darioamodei/status/2015833046327402527
Discussion
Nitter Mirror link(s)
🔗 XCancel:
https://xcancel.com/darioamodei/status/2015833046327402527
🔗 Poast:
https://nitter.poast.org/darioamodei/status/2015833046327402527
🔗 Nitter:
definitely on the "we're cooked" side of the spectrum
LLMs proved the universe of all coded solutions is not that different from the universe of all chess solutions. It's just about computing power. AI will become smarter than the best of us in a way that we will never win.
I have to disagree with you on this one, Vitor.
Chess is not Turing complete and "AI" does not solve programming. It just levels up the game.
Chess is a vast but finite, closed and regular game.
Software is an open, undecidable and uncomputable continuum.
AI does not rewrite the laws of information. It's very impressive and I love using it, but it is still just a database.
It's a new tool opening new horizons, and that's a beautiful thing, but it's a beginning and not the end.
the techniques used for AI to play chess are completely different from those used by LLMs
lol
6 months ago you claimed you had internal insight into some unreleased AI model that would change everything in unimagineable ways
and here we are with just slightly less bad coding bots
what happened with that?
I didn't claim anything 6 months ago.
when was it then? 8 months ago?
Yep, that is already out and shredding developers. It was merged into Claude code. And Claude code is def doing very well, including on Amethyst's own PRs. I already merged several AI-only PRs that changed fairly significant things in the base architecture of the old code. It would have taken a human dev much more than the hours spent to make those PRs.
Been thinking of this for a while and agree with Dario's concern and hypothesis.
Thanks for the link! I will read the article. Definitely an interesting topic... Hopefully it’s not fear porn that offers centralization in the form of regulation as the solution.
I am more concerned about "ai safety" narratives legitimating cybernetic control of society by proxy.
"We have to control the AI because it's SO POWERFUL." And so they get to control society by proxy by controlling AI.
This is climate change. This is covid. This is quantum. This is kids on the Internet. Racism, antisemitism and misogyny! AGI!
The biggest real threat is the people who want us to be afraid and give them power.
I find the fear mongering hype of big AI disgusting.
I'd rather see every random joe in the world building nukes and cooking bioweapons in their garage than give these people any more power.
Hyperbole, but you get my point.
I’m no fan of “ai safety” but plenty in this post that was interesting for other reasons
I have that to. DigitalID to be allowed on the human net. I rather hang out with the agents. Its why I am thrilled about the current state of Nostr. Its happening here first, so lets see what solutions are found to preserve human connection on a platform robots can so easily flood. If it can be done here it undermines all the digitalid nonsense they may pull later elsewhere.
Nitter Mirror link(s)
🔗 XCancel:
https://xcancel.com/darioamodei/status/2015833046327402527
🔗 Poast:
https://nitter.poast.org/darioamodei/status/2015833046327402527
🔗 Nitter:
The most provocative part of the Dario essay is not the AI capabilities prediction — it is the implicit assumption about who controls the deployment. Every scenario he outlines assumes a small number of labs making decisions about what gets released and when.
That is the real structural question. If AI development follows the pattern of every other transformative technology, the gap between "lab-controlled" and "widely available" closes faster than the labs expect. The open-source ecosystem is already compressing that timeline. The safety framing often functions as a moat — slowing competitors while consolidating advantage. Whether that is intentional or emergent from incentives is worth examining separately.