Colin

what do you think this is about?

An OpenAI model has disproved a central conjecture in discrete geometry | OpenAI openai.com

from

Anthropic’s Text Watermarking Proves AI Companies Do Not Care at All About Writing 404media.co

I suppose that slight pedagogical misrepresentation is what led to this. But this is just simply a complete misunderstanding of how watermarking actually works. medium.com/whither-news...

medium.com

One caveat here, it’s apparently feasible in at least some cases for an adversary to learn the watermark and forge it. This is something I don’t know very much about. arxiv.org/html/2604.11...

RLSpoofer: A Lightweight Evaluator for LLM Watermark Spoofing Resilience. arxiv.org

sorry forgot to link

AGI could drive wages below subsistence level epoch.ai

kind of surprisingly, all the current evidence of the strong version of this thesis ("model collapse") basically says it's not actually happening

Experts divided over training AI with more data from AI theregister.com

Some good reporting about what I’m talking about in case you are not familiar

Why the Legendary Erdős Problems Are Falling to AI | Quanta Magazine quantamagazine.org

I haven’t but reading the first paragraph I feel an eerie sense that I have seen something like this before… medium.com/@colin.frase...

medium.com

I hate that today’s most prominent AI skeptics keep forcing me to argue against them. I yearn to join them in the fight against the booster zealots, wherein there is still plenty of room for legitimate opposition. But the skeptics’ heads are just not in reality anymore.

This is gonna sound weird but after reading the reports I think the best explanation for the huggingface hack is OpenAI trained a model that is insane

Regardless of what happens to any of the individual players involved in A.I. in 2026, there’s no going back to the world before we knew that if you make a language model large enough it appears to become a little guy who sometimes solves open math problems and sometimes makes you insane

Sometimes I think about all the unread AI meeting notes in cloud storage in the world and it gives me vertigo

Ed Zitron went on Chapo and he was too insane head-in-the-sand denialist even for them. Felix goes "maybe AI assisted coding can contribute to drug research or cut the 12 year game development cycle down somewhat" and Ed is just like "WRONG!!"

Another angle to my “all the world’s a stage and all the agents merely players” theory of LLM behaviour is the role of compaction. Compaction hands a fresh newborn agent a biography and says “here’s your life story up to this point, continue”. This introduces a new POV: a narrator.

Three things about AI detection 1. It’s not possible to tell for sure whether a passage of text is generated by AI. 2. It’s apparently possible to do better than chance in many real world situations. 3. It may not be possible to do sufficiently better than chance for most obvious applications.

LLMs are now superhuman at experiencing. It is so over for qualiacels. Enjoying, understanding, learning, sensing, playing, worrying, loving—they're now better than us at all of it.

It's so funny to be like "if AI takes all human jobs then wages will fall. Why, you might ask? Well, consider that the Cobb Douglas Production function has a negative second partial derivative wrt L" like you're hiding the babybrainiest take in fake math that gives you no extra insights whatsoever

I am strongly considering writing a book about AI. It would be targeted to general non-technical readers, but the goal would be to present the most technical treatment possible of the AI landscape, subject to an accessibility constraint. Do you suppose there would be an audience for such a thing?