magi🌙

22, CS student & budding writer, pfp by @shirumidoriiro | github:

[magi🌙 @hotlinemagi]

@purnya2 OH

[magi🌙 @hotlinemagi]

@rtiaho computah

[magi🌙 @hotlinemagi]

@0xThoughtVector there’s a reason why they were complicated for the general public by the way

[magi🌙 @hotlinemagi]

I have been repeating this for WEEKS. That’s what happens when you study.

[Valerio Capraro @ValerioCapraro]

BREAKING: OpenAI’s solution to Navier–Stokes does not match its Lean verification.

The most important article to read today is not one of OpenAI’s 700 AI-generated math papers.

It is this other paper, making a deep and worrying point:

A Lean-verified proof does not automatically validate the proof written in natural language, nor does it mean that the formal statement captures the intended theorem.

During translation, an AI can change an assumption, weaken a statement, or replace the argument entirely.

It can hallucinate another theorem.

Lean correctly verifies the result.
But the proved result may no longer be what the paper claims.

This is a general problem. Things get spicy when the authors examine OpenAI’s proposed Navier–Stokes solution.

They identify at least two mismatches between the written intermediate results and their Lean counterparts:

One estimate claims that four additional input derivatives suffice. The Lean version requires five: a weaker result.

A pressure-flux estimate is obtained through a different bound, and proved through a different argument.

It is not clear whether these mismatches invalidate the entire proof.

But they raise an important issue.

OpenAI is flooding us with claimed revolutionary breakthroughs. Yet nobody knows whether the proofs are correct or whether they prove what they claim to be proving.

Epistemia at scale.

*
Paper in the first reply

[magi🌙 @hotlinemagi]

@olivers_tools This is not something new btw. There’s a video of like 4 months ago where its done with 100 players.

[purnya @purnya2]

a simple study z

[magi🌙 @hotlinemagi]

Pre-print and proof to follow.

[magi🌙 @hotlinemagi]

Tightening OpenAI #109 to the global optimum: κ = 1.

By replacing numerical branch-and-bound heuristics with a canonical algebraic certificate, we bypass the remaining ~2.38 × 10⁵-fold gap left by the 4.19 × 10⁻⁶ witness, closing the deficit from 2⁻¹⁸ to 2⁰.

[Swapnil @SJ_Swapnil_Jain]

We are publishing a third update to OpenAI problem #109 (integer multiplication), and we have now entered 2^-1x territory.

κ > 2⁻¹⁸ (tightened from κ = 2⁻¹⁸²)

The exact witness is 4.19 × 10⁻⁶, roughly 558 fold over our previous 7.5 × 10⁻⁹, about 2.2 fold over the strongest open result, and a 2¹⁶⁴ fold improvement over the original OAI result.

[Shamaboy @Shamaboy11]

Good-ass video that resonates with me so much

[magi🌙 @hotlinemagi]

@iD7M19 best game in existence

[magi🌙 @hotlinemagi]

@dysmemic It’s linear or it’s nothing🙏🏻

[magi🌙 @hotlinemagi]

@omeedtehrani @constspace why only US :(

[magi🌙 @hotlinemagi]

oh yeah also @opencode

[magi🌙 @hotlinemagi]

Rex by @superlogical and @pidotdev by earendil are - for now - the only high quality and well thought software post AI

[magi🌙 @hotlinemagi]

@almonk I want them

[magi🌙 @hotlinemagi]

@theo @bta_ckr *sighs*

[magi🌙 @hotlinemagi]

AI has killed almost all my interest in other’s projects.

[magi🌙 @hotlinemagi]

Maybe the future is gonna be superhumans playing God with Life simulators.

[magi🌙 @hotlinemagi]

@mgostIH nobody will actually use this

[magi🌙 @hotlinemagi]

@dmiiiitrii ITS SO CUTE