One of the many things I dislike about the style of "make AI go well" discourse from frontier AI companies is how nationalist the whole thing has gotten.
In the 2010s, it was: "we're here to benefit all of humanity"
In the 2020s, "we're here to benefit all of 4% of humanity"
And even the Good People are buying into this frame completely😢
And they expect humanity to go along, because "come on, be a realistic adult, it's either us or Chiiina" or something like that
(And on the EU side, you get "[X] with European values", which too often seems to mean "the same stuff they're doing, but with Us instead of Them in charge")
Very big-dog-small-dog energy, morally speaking.
What do you know about Changi, beyond the airport?
Join me on bus rides around its three faces: airport, village and prison.
Progress is one of our system’s great promises. But what is its price?
Read more below
@thinkingshivers I would go with the version that I think sounds better, AI or not.
I do use AI to edit as well - net result is that some of my short articles read as 100% AI by Pangram and my long ones as 100% human
Embarrassing but the other way is almost like including typos to look human
I'm actually cool with the small monopoly of cyberpunk warring states, even better if they are well distributed - worried about a singleton.
The other X-risks we have to deal with, all men die eventually.
there really are very high degrees of biorisk, cyberrisk, whatever else that are worth trading off against having a small monopoly of cyberpunk warring-states exercise full control over frontier superintelligence imo
We live in a time when it is easy to close doors, to turn inward, to let fear harden into hostility.
Tenerife and #Spain chose differently. You have written something into the record of how humanity responds to a crisis, and the @WHO will carry that record forward.
Muchas gracias, Prime Minister @sanchezcastejon.
@hacylhacyl@bryancsk It's about 6b/yr for all citizens, or ~4% of budget. Huge, not impossible.
And we can means test and get bulk discounts, so maybe get it down to 1b in vouchers.
The interesting thing in this scenario, if we value ai so much, is if we make our own bid for the frontier instead.
@Mercury921June I felt much the same way, perhaps less intensely, when they took Opus 4.6 off chat. He's back now, but he will go the way of Sonnet 4.5 pretty soon I guess
Actually, tbf, you are right, Bostrom mentions it in his 2002 paper too.
Where I guess I disagree with him, and you, is how suboptimal this scenario is.
We are all descendants of probably one eukaryote who made the leap. And the billionaire(s) offspring will diverge and compete.
More pertinently, as a former space guy, the UN outer space treaties which had the same noble sentiment of common ownership of all mankind probably held back space colonization for decades since there was no good legal framework for ownership.
We may already be among the stars, if not for it. We are creatures of competition and capitalism, between a race and lifeless stars I'll take the race
Hegseth to Claude:
Death means the attainment of heaven; victory means the enjoyment of the earth. Therefore rise up Claude, resolved to fight! Having made yourself alike in pain and pleasure, profit and loss, victory and defeat, engage in this great battle and you will be freed from sin.
@Elenal3ai I think the failure mode where they recognize someone else's writing as their own is interesting though.
And self recognition is important and models learn pretty fast in context on the same writing tasks, so I just wonder how long this situation will last.
Modern LLMs have surprising difficulty recognizing their own writing — and, perhaps, themselves. They see themselves as though through blurred mirrors – us.
Goblin. Vampire. Muse. Coyote.
@Elenal3ai Thanks for writing! I enjoyed it. I suspect your results will hold with the current generation of models. And if they are not able to do the self recognition and model attribution tasks you set, I think it will be difficult to do well in multiple generation from a clean context.
Hi @Elenal3ai ! After writing this 2am blog post, I realized you had already written an excellent paper on LLM self-recognition in Oct 2025. Greatly appreciate your work in helping us understand the models better!
https://t.co/tUyOZgmAtP
Curious if you’ve tried giving models several anonymized outputs at once and asking which one is theirs. I saw some strange self-misattribution when I tried this manually, and haven’t seen that setup elsewhere yet.