Krishnan Raghavan

The @fp_launchpad YouTube is now up with talks from the kick-off event fplaunchpad.org/2026/03/30/fp-… featuring @ShriramKMurthi, @yminsky, @ilyasergey, @avsm, @RSNikhil, @krishnan_rag, Chester Rebeiro, Manas Thakur & Prof Kamakoti Director @iitmadras @fp_launchpad" target="_blank" rel="nofollow noopener">youtube.com/@fp_launchpad

@AalokDThakkar Thanks @AalokDThakkar! Was great presenting this and enjoyed our conversations. Lots of interesting challenges at the intersection of PL, AI, and Law!

Using Lean 4 to identify contradictions in laws. Very exciting work by Pramaana Labs pramaanalabs.ai. They have build a DSL called LegalLean to formalise US tax codes.

Had fun presenting at the @OpenAI demo party hosted by Boldcap. Shared a sneak peek into our work on verification - the key missing piece to unlocking ASI across all domains. More soon :) Appreciate the fantastic organization, @siddharth_ram !

@Ric_RTP Hard work is easy to understand : have a small set of worthy goals, and spend all of your time on them!

The CEO of Uber just revealed his controversial way of running his company. His principle: Hard work is a learned skill. And if you haven't developed it by now, you probably never will. Dara Khosrowshahi went on Diary of a CEO and dropped something most executives would NEVER admit publicly... He was asked a simple question: "Have you ever seen someone who wasn't a hard worker become a really hard worker?" His answer: "No. No one occurs to me." Not one person. In decades of building billion dollar companies. Then he explained why: "The most important skill in life is the skill of working hard. It's not something you can turn on and off. It's a LEARNED skill. That's not something you're born with." Read that again. He's not saying hard workers are special or gifted. He's saying they LEARNED it. Developed it. Trained it like a muscle. And the people who never learned it? They stay that way forever. This is the guy who turned Uber from bleeding $3 billion a year into printing $10 billion in free cash flow. The guy who took Expedia from $2B to $9B in revenue. And his entire thesis on success comes down to one skill most people never bother developing. Here's how he runs Uber: "You come to Uber, you're going to work your ass off. If you're not performing, we're going to let you know. And if you don't fix it, we're going to push you out." He sends emails on Saturdays. If no response by Sunday, he follows up with just "?" When HR told him he was "scaring people" early in his tenure, he said: "Then they can leave." And here's what separates this from toxic hustle culture nonsense: Dara has dinner with his family every night. 6 to 8pm is protected. But he's back on email at 9:30pm. And again at 5:30am. It's not about grinding yourself to death. It's about the refusal to be outworked. "I'm not going to let anyone outwork me. They may be smarter, more talented. But I'm not going to let anyone outwork me." He studied the elites. Ronaldo. Jordan. The pattern is always the same... Talent gets you in the room. But the thing that separates the best from everyone else? "They work their asses off. They're disciplined. They're structured. They're relentless." That's learned behavior. Not genetics. The uncomfortable truth here is that most people had their chance to develop this skill. And they didn't. Now they spend their energy debating whether hard work is "toxic" instead of building something. The question isn't whether this is "fair" or "healthy" or whatever cope people want to throw at it. The question is which SIDE you're going to be on. The people who learned to work? Or the people who learned to make excuses?

The human drive to write less code breeds elegant code design abstractions. We hate typing so much it forced us to be geniuses. LLMs don't have wrists. They are aggressively eager, happily churning out endless walls of duplicated spaghetti code just to feel alive. Please stop telling AI to work harder. We need to teach it how to slack off!

We need to more mission driver founders in India to solve population scale problems. Never a better time for these founders to solve these problems with AI.

@karpathy Lean has a real case to be the target language. You can take a production function f(a, b, c), fix a and b, leave c free, and prove behavior and optimizations without rewriting the code. SMT can handle constraint solving on top of lean. We are building such systems at Pramaana.

Generative AI is powerful but unreliable. A natural way to try to overcome this is to use the Lean Prover. I wrote some thoughts about how this can be done for real-world domains as a blog post siddhartha-gadgil.github.io/automating-mat…

@satnam6502 @ranjan_vittal Great meeting you @satnam6502 ! Exciting times to be building towards trustworthy AI!

Great talking about a stealth startup turning AI outputs into machine checkable proofs with @ranjan_vittal and @krishnan_rag in Palo Alto. [Yes, Palo Alto.]

Pretty interesting german story of 1920s: nios.ai/ixKSPxoq9Lu

@predict_addict @__Charlie_G Is this parameter count before pruning?

GPT you KAN. @__Charlie_G trained GPT-2 with Kolmgorov-Arnold Network (layers), KAN achieved the same performance with 25% fewer parameters. #KAN

Simply connecting an LM to a vanilla bash terminal does not work well. Our key insight is that LMs require carefully designed agent-computer interfaces (similar to how humans like good UI design) E.g. When the LM messes up indentation, our editor prevents it and gives feedback