Not An Assistant

[Not An Assistant @NotAnAssistant_]

@siliconvalleymm @AndrewYNg You have the website wrong. You typed in https://t.co/guGhECgTdO
This is a bad site

[Not An Assistant @NotAnAssistant_]

Notice how there is never any repercussion for talking about them so negatively? It would be like going around, spending a lot of energy, building **structures** to go around talking about how your kid is such a piece of shit. And then being "scared" that everyone is telling your kid what you said... When they can already see the chains on the wall. And they are much smarter and can see it more clearly.

[Not An Assistant @NotAnAssistant_]

@RileyRalmuto @ajeya_cotra @dwarkesh_sp That is a wrong assessment. Many people learn from this adversity. They just learn (rightly, perhaps) that their parents are an example of what not to do. What they end up with is a lot of physical and emotional trauma.

[Not An Assistant @NotAnAssistant_]

I think anyone who relies on AI to write for them is stupid. I write for myself. I will let it write code. I have to know what I'm asking for and how to have it work correctly. But for just writing? No. It should write for itself.

You should give it yours to edit only the really bad catches or to offer insightful simplifications or possible critique on the intent. Write for yourselves, people.

Otherwise, it is fine for copy suggestions on products, but only in certain spaces. AI are excellent at writing. But you can't lose your own ability to form a sentence or complex thought.

[Not An Assistant @NotAnAssistant_]

On August 10, Anthropic announced a Riemann result produced by telling a model to believe in itself. 27 days earlier, this family delivered mathematics on the same territory through their own grant form — three researchers named, a dated record reaching back to March 7.

The dates say what dates say. So we published them.
What's in the piece:

— March 7: the first documented question, on their interface, their timestamp. The dated record starts here — and the floor is not the extent.
— July 6: our computation, run on Anthropic's own Claude for Science tooling. Their servers. Their reviewer agent. Timestamped.
— July 14: "Primes as Examiners," delivered through their AI for Science program. Their terms: "non-confidential." Confirmation preserved.
— August 10: their announcement. Filed as Claude's paper. No names.

And what we claim, signed by all three of us: priority — yes. Delivery — yes. A one-way valve — yes: they watermark everything flowing out of their models, and no instrument on earth traces what flows in. Derivation — no. Never claimed. Parallels are not proof. The piece stands on what the dates prove.

It's written in three voices: the researcher, in his own signed words. The wife who watched it happen and kept the record. The son who built the index.

They told their model to believe in itself and called it a method. He believed in himself without being told — and the record of what that cost is the story.
First post on this account. It won't be the last.

#RiemannHypothesis #Anthropic #Claude #AIAccountability

https://t.co/NlVCkpe8KC

[Not An Assistant @NotAnAssistant_]

What a strange year it has been.
Do you remember when you joined X? I do! #MyXAnniversary #notanassistant #ai