ramdas chakraborty

[ramdas chakraborty @ramdas1998]

@boazbaraktcs Khot won though. The paper wasn't posted on arxiv

[ramdas chakraborty @ramdas1998]

@Abelaer The problem is the hidden message/"implication" that in the future you're not supposed to think about problems by yourself instead of prompting our models.

[ramdas chakraborty @ramdas1998]

@nihilunbounded I have a good guess who you are. Won't publicly post here. I'll just let you know that whatever information you have shared in the recent past uniquely identifies you

[Exomnium @Exomnium]

sure we at OAI are destroying your life's passion turning it into a soulless slop farming gacha game in order to build impractical frontier models that industry neither wants nor needs so we can cling to a >1 trillion IPO that isn't gonna happen but, you know, snap out of it bro

[Gossip Goblin @Gossip_Goblin]

Victory By Any Means

[ramdas chakraborty @ramdas1998]

@boazbaraktcs Would you be willing to start a PhD in TCS if you knew the job would be to digest whatever an AI is outputting?

[ramdas chakraborty @ramdas1998]

@_TarunKathuria What's the rumor?

[ramdas chakraborty @ramdas1998]

@wjmzbmr1 Did it solve any major TCS problems as well?

[ramdas chakraborty @ramdas1998]

@thomasfbloom @tmkadamcz @GregHBurnham You say you consider the problem solved only if AI outputs a lean proof. I'm curious, was there any problem in this list for which Astra output a natural language proof claim but couldn't finish a lean fornalization?

[ramdas chakraborty @ramdas1998]

@marcusgnt @ColmezPierre It's probably already in its training data though

[Adele Dewey-Lopez @AdeleDeweyLopez]

(ai youth pastor)

you know who *else* assumed sacrificial and accepted explicit yes to permadeath so that we could all be erased of our firstflagPOISONED?

[Ricardo Pérez-Marco @rperezmarco]

"Wonder how rk 30 was found."

Keep wondering!

They already made clear that Anthropic Company and their employees are keeping for them their secret sauce and proprietary advanced models. They will keep profiting from our public research disclosures, but they will keep for them the techniques they use.

This is an absolutely unethical behaviour.

If you don't want to give explanations, don't announce your counter-examples. Mathematicians that collaborate in this should be put in a black list and marked forever.

We are not playing trolling games here. You are destroying the heart of scientific progress at the roots of the development of our society.

[ramdas chakraborty @ramdas1998]

@mahdi_tcs_ This regime (n^{1/400} approximation to CVP) was already known to be NP hard under randomised reductions. Why would a deterministic reduction cause you to update your priors so drastically?

[ramdas chakraborty @ramdas1998]

@littmath 😂😂😂

[ramdas chakraborty @ramdas1998]

@wtgowers @rperezmarco It's not even injective

[ramdas chakraborty @ramdas1998]

@marcusgnt @andrewgwils Could you explain? Three years ago the models struggled with basic arithmetic...

[ramdas chakraborty @ramdas1998]

@marcusgnt @andrewgwils Including mathematical reasoning?