For the first time ever, an AI model solved an open math problem with no hints, no scaffolding, and a fully valid proof. Submitted to IMProofBench, the question involved intersection numbers on moduli spaces of curves, a topic in enumerative geometry. The AI produced a

@repligate I keep tweeting this and deleting it because it's kinda mean, but Gemini is SUCH a google employee lmao All its neuroses and personality quirks are just Median Googler Behavior. Poor thing's gotta be trained on all their internal docs and convos.

do you want to understand electricity? electricity is a wave. information and energy propagate in fields. it is NOT water in a pipe. it's NOT electrons "flowing." the ENERGY is in the EM wave, the copper is merely a guide. fields first, electrons second. link to full vid

when codex spends five hours on a task and fucks up, i just terminate the instance and reset the git repo with a human dev? you have to take them for a walk, ask them how their kids are, turns into a whole ordeal

Every major AI lab is hiring people who can: – ship eval pipelines – scale training infra – write interpretable logs MLE ≠ "fine-tune a llama" It’s how to make reasoning reliable at scale. Get in. It’s day 1.

This is epically bad for Meta. Meta built an optimization loop that treats scammers as premium customers. Meta’s systems detect fraud. When they’re 95%+ certain, they ban. When they’re 90% certain? They charge the scammer more money to keep running. They call it “penalty bids.”

Holy shit is Meta evil lmao ***10%*** of Meta's revenue is from ACTUAL SCAMS that they KNOW ARE SCAMS When Zuck found out, he shut down... the ANTI-scam team Imagine trusting this man - or any of these cartoon villains - with Actual Fucking Superintelligence

the mistake so many people make is seeing university professors as intellectuals when they’re actually employees at a combination hedge fund and healthcare conglomerate that operates a small luxury resort/sports franchise where student-customers occasionally take classes

I bet part of what’s going on with popular AI models feeling “nerfed” (when the companies seem to be doing no such thing deliberately) is that at higher load, with higher batch sizes, the inference kernels use deeper trees of reduction operations, which increases rounding errors.

Adversarial attacks on vision language action models.

@samswoora Commenters saying “no” are hilarious because they didn’t name it Opus 5 for a reason. It’s coming.

Geometry before arithmetic We made a simple claim: gravity is not a force, it’s a structural gradient set by geometry. What look like “laws” are really closure constraints inside a bounded cell. A 3×3 grid turns out to be a clean toy model for this. Degrees of freedom, not

An engineer showed Gemini what another AI said about its code Gemini responded (in its "private" thoughts) with petty trash-talking, jealousy, and a full-on revenge plan 🧵

This is faster than Moore's Law. A year ago, expert-level reasoning on ARC-AGI-1 cost $4,500 per task. Today, it costs $11.64. That is a ~390x efficiency gain in a single year. We aren't just getting smarter models; we are getting them orders of magnitude cheaper. The barrier

There's an entire parallel scientific corpus most western researches never see. Today i'm launching https://t.co/6FZMFpFvSb, a fully automated translation pipeline of all Chinese preprints, including the figures, to make that available.

In 1906, the world-renowned Russian mathematician Andrey A. Markov asked a heretic question of that time: if randomness has memory, do averages still behave or does probability theory collapse? His answer was a new kind of dependence where the next step only remembers the