Associate Professor at Georgia Tech and University of Toronto (On Leave). https://xitter.cf/t.co/Cm8BZMIOqc Automated Reasoning | Formal Methods + AI
Toronto
Joined May 2010
- Tweets2.3K
- Following321
- Followers2K
- Likes3.1K
Pinned Tweet
At today's Stanford theory seminar, Don Knuth talked about our algorithm for distinct elements estimation in a stream.
Picture credits: Harshit Mittal.
And here is the note that he posted: cs.stanford.edu/~knuth/paper…
That dinner with Don (Knuth) changed everything. 1/4
Kuldeep S. Meel retweeted
One heartening thing about the @AnthropicAI + @theoremlabs verification work: almost every PL researcher on my feed is excited about the future and seeing years of research put to use. Quite a contrast to the “AI is taking our jobs” mood I’ve seen in other communities.
Kuldeep S. Meel retweeted
William Dalrymple has recently shown himself to be a giant of moral integrity and intellectual courage—unlike the many weaker intellects who shill for a (dying) establishment.
Just amazing. Never thought that I would get to see (qualitative) formal verification possible at this scale.
Can't wait to see how we go about quantitative reasoning and designing systems correct by construction.
The teen years of CS are over!
New on our Frontier Red Team blog: using Claude to formally verify software at large scale.
Formal verification can prove that code is free of certain bugs, but historically it has only been feasible for small codebases (or required a lot of expert time).
With AI models, however, this is beginning to change. In collaboration with @theoremlabs, we used Claude to prove security properties in over one hundred real-world open-source libraries over the span of a few weeks.
Read our research in full here: anthropic.com/research/forma…
Academia's obsession with reference letters was always unscientific and thanks to LLMs, we have a chance to move past it. Without LLMs, it was never going to happen as this was the ultimate gatekeeping.
Because part of the reference letter is costly signaling - that a busy person is taking time out to think deeply about why they recommend a particular student. That signal is now gone - dare I say even negative.
Second - I’m just tired of Claude speak. Can spot it a mile away and so the letter is less distinguished from all others that sound the same.
This is important enough that I’m paying attention to substance - but the tradeoffs are not worth it IMO!
Letter writer is senior faculty and native English speaker so many traditional reasons why this would be ok also don’t apply.
Kuldeep S. Meel retweeted
I’ve been released from IDF custody with a “warning,” apparently because of my public profile (I told them I wasn’t even a congressmember yet, just a nominee).
But several dozen rabbis & participants harvesting olives with me are still in detention. For harvesting olives. They should all be released immediately.
To be clear: Palestinians face far worse consequences every day. These detentions by the IDF are one more attempt to undermine Palestinians’ right to harvest their olive groves, travel on their roads, and live in their homes. Yet one more element of escalating, and very clearly state-sponsored ethnic cleansing.
We’ll update on their status as soon as we hear.
Kuldeep S. Meel retweeted
A heartwarming viral video shared by Hyderabad businessman Nishant Saboo captured a sweet gesture at a luxury car showroom. After spotting two uniformed schoolchildren admiring a Lamborghini from outside, the owner warmly invited them inside, letting staff click their memorable pictures beside the supercars and winning hearts across social media.
A major challenge for auto-formalization is how to formalize at a level of abstraction that are reusable.
Lean, IMO, is too low level: Look no further than OpenAI's Lean repo: they proved everything in every formalization from scratch.
Two days ago, I had Tex2Lean to have a go at "An FPRAS for Cell-Bounded Contingency Tables", and we now have a Pseudocode and associated Lean Proof. (Link below)
[OpenAI's Lean proof had a Turing Machine description, which is too low-level unless you are a Turing Machine].
Kuldeep S. Meel retweeted
Ori Goldberg was a thoroughly decent man who lived the “yes, you’re all wrong” meme by standing up to his countrymen every day for the last 3 years. I know it wasn’t easy but he never complained and his courage inspired countless others, including me.
We now have formalized pseudocode for counting common bases of two matroids.
xitter.cf/ksmeel/status/21090177…
I put Tex2Lean to extract and formalize the algorithm presented in the OpenAI's paper on counting common bases of matroids, and 48 hours, we have a formal pseudocode whose correctness is proven. 1/
There are some proof steps that preprint claims to need but there doesn't seem to be a one to one correspondence in the lean. For example, preprint seems to require "The statements used here are cite[Theorem~4.10, Theorem~2.10, andCorollary~2.11]{BH2020}, 3/
Kuldeep S. Meel retweeted
The people who think that LLMs are producing all these proofs by stealing the work of mathematicians who are using their services are deluding themselves. The sooner we accept how powerful this technology is, the better able we will be to deal with its consequences rationally.
I am not surprised;
This is what I warned yesterday: xitter.cf/ksmeel/status/21076397…
I believe Tex2Lean infrastructure can help!
The plot thickens.
"We show that the formalised Lean proof does not correspond to the [written paper] proof of blow-up of solutions to the Navier-Stokes equations."
arxiv.org/abs/2610.08144
Kuldeep S. Meel retweeted
It sounds insane, because we're used to a particular way of doing research: of giving and receiving credit, of counting beans, of hiring, recruiting, and promoting. And the machine god came to crush that game. The old game is dead. That reasonably causes incredible amounts of dread for many folks, who understand this as the death of research.
But it's not about Research, but the way we played the game. The old game was built around scarcity, scarce compute, scarce students, scarce hours, scarce people who knew the right technique and method etc. Most of a research career went into managing and playing along these constraints. Those walls just came crushing down. One person with a good question can now do what used to take a lab and five years.
You do not need permission, you do not need a big grant, you do not need a professor or researcher title, or a team of twenty experts , or one thousand gpus to stare straight into the first light of the universe.
How absolutely incredible
