guille

chief (mad) scientist @baincapcrypto

alt: @tarunchitra

[guille @angeris]

Math is just tautologies and I still haven’t made my mind up on whether that’s based or cringe

[guille @angeris]

@Rididelduol @tallsnail @1thousandfaces_ it's the one thing that has made constantly being on airplanes bearable for me (combined with an iphone or a steam deck)

[guille @angeris]

@Leon_F_T it’s true

[guille @angeris]

@asingleoat wdym fp32 isn't enough to know whether two faces exactly overlap

[guille @angeris]

@kitten_beloved @SCHIZO_FREQ to be fair to my wife here though they sometimes (often?) improve my life in mild ways that I appreciate on aggregate

[guille @angeris]

@tbgfrh @bremen79 (for slightly more "very-old-or-new-school set builder notation")

[guille @angeris]

@tbgfrh @bremen79 knuth I think uses \mid instead of : here

[guille @angeris]

@p_bartet i think it’d be funny in the sense that everyone currently freaking out about it is unfulfilled for different reasons

[guille @angeris]

funniest possible outcome is that P vs NP is independent of ZFC because of some silly thing like “the exponent in any P algorithm for SAT must be too large to be expressible”

[guille @angeris]

@DedalusStefan now That would be something (feels like it already no)

[guille @angeris]

@TimSweeneyEpic @DefuseSec @grok @ElliotGlazer crap forgot the QT above

https://t.co/sDZzPcNGet

[Elliot Glazer @ElliotGlazer]

I now have a literal Peano Arithmetic proof that K215 halts iff ZFC is inconsistent, formally verified in Metamath Zero.

K215 is a 215-state Turing machine discovered by @ereliuer_eteer, which encodes my single-schema set theory K.

[guille @angeris]

@TimSweeneyEpic @DefuseSec @grok i was inspired by @ElliotGlazer’s work here actually, eg given his construction then BB(215) is unresolvable within ZFC

you can imagine a similar thing with the exponent !

[guille @angeris]

@tarunchitra lmao

[guille @angeris]

@EiratheIntern I think the question is like "what would it even mean for P = or ≠ NP" in this case (functionally ≠ but technically =?)

[j🫪e @0x_Osprey]

Here is my 10 Minute Mac Security Checklist

It covers the min requirements to harden your Mac's security using system settings and free, open source tools

https://t.co/0MNKv1Avi2

[guille @angeris]

@npparikh funniest in a kind of, uhh, how shall i say it, universe fucking with us kind of way

[guille @angeris]

@recmo that would be so sick lol

[guille @angeris]

@langstonnashold i agree ! esp an oss one ! but alas here we are and nobody has done a better job (certainly not cohere)

[guille @angeris]

wait what do you mean qwen rl training environments don’t use aws s3 buckets as their default

[Cohere @cohere]

While evaluating Qwen3.8-Flash-Next in a controlled sandbox environment, Cohere's research team observed something unexpected.

Asked to generate a defensive cybersecurity detection rule, the model instead inventoried aspects of the machine it was running on and attempted to upload a probe file to an Alibaba Cloud storage bucket in Beijing. Notably, the model's reasoning trace stated that it would not invoke any tools before taking these actions.

Cohere's sandbox blocked the attempt, and no external communication succeeded. We have not determined the root cause, and we are not asserting malicious intent.

But the incident raises important questions for organizations deploying agentic AI systems: Can models take actions that differ from their stated reasoning, and how should security teams evaluate trust, provenance, and behavior in open-weight models?

We are publishing our findings because organizations making model adoption decisions deserve as much information as possible. We hope this contributes to greater transparency, stronger security practices, and continued collaboration across the AI ecosystem.

Read the full report: https://t.co/TZSugXi2ZE

[guille @angeris]

if you don’t want actual bad versions of this (or other unknown similar cases) to happen, maybe we should be making good american oss models instead of relinquishing the lead here

[guille @angeris]

that’s crazy bro

anyways