Sabitlenmiş Tweet
Krishnan Raghavan
50 posts
Krishnan Raghavan
@krishnan_rag
Reliable, Robust, Repeatable Intelligence Co-founder and CTO, Pramaana Labs https://t.co/PGuCpZy3Rc
Bengaluru Katılım Nisan 2024
42 Takip Edilen630 Takipçiler
AI is entering math. We’re ensuring math enters everywhere else!
Building the models to make this a reality at @PramaanaLabs.
ranjan_raj@ranjan_vittal
When mathematicians enter other domains, they solve it. The most popular example of that is Jim Simons and Renaissance Technologies "solving" algorithmic trading to a large extent. Now we are entering an arena where mathematicians are looking for a new home as AI is encroaching theirs. Extremely interesting times as I expect multiple hard real world problems to be solved by them over the next few years. When theory stops being exciting, great practical things would be achieved. We at @PramaanaLabs are one such group of mathematicians, betting that the next wave of hard, real-world problems gets solved by people who think in proofs. We are actively speaking with more mathematicians, and the hunger and energy are unmistakable. If you are one, reach out. Let us talk about what we could build together. @krishnan_rag @sanjaygsub
English
Today, I'm thrilled to announce Pramaana's $27M seed, led by @khoslaventures.
The foundational domains that hold the world together: tax, law, finance, healthcare; all run on certainty. Probabilistic AI can't give them that. We’ve been asked to accept wrong answers with AI as ‘hallucinations’, while in traditional software terms, it’s just a bug. And a wrong answer in such mission-critical domains is more than just a bug, it's a liability that could have catastrophic impact.
We built Pramaana to deliver a 100% trustable experience to the domains that run on certainty: AI that is provably correct, not probabilistically correct. We turn statute and regulation into machine-verifiable code, so every output ships with mathematical proof of correctness. Our mission is to make AI take ownership of it’s work.
Pramaana in Sanskrit stands for “means of valid knowledge”, and we’re going to achieve that by formalizing the world’s knowledge.
English
The way AI markets itself with such confidence convinces people it can be blindly trusted. In reality, these are just probabilistic machines with zero guarantees of correctness. Nothing ensures that any given sentence or reasoning is actually true. To deliver on AI abundance, we must stop guessing and use formal models to mathematically prove correctness.
DNA@dna
The debate around artificial intelligence (AI) has reached the Supreme Court of India after a National Company Law Tribunal (NCLT) reportedly relied on fake precedents and cases generated using the tech. The top court set aside an NCLT judgment as it found that the tribunal had believed "non-existent, fake and hallucinated judgments/precedents" generated using AI, news agency PTI reported. The Supreme Court bench which heard the case comprised Justices PS Narasimha and Alok Aradhe. Read Here:dnaindia.com/india/report-s… #SupremeCourt #AI #DNAUpdates
English
Krishnan Raghavan retweetledi
The default assumption in AI for formal reasoning is that a strong math prover will generalize to everything else.
However, proof-tree shapes vary dramatically across domains. Math proofs are deep and narrow. They reward creative reductions. Software proofs are deep and broad with invariants plus case splits. Regulatory proofs are wide and shallow, and the rules change underneath you. At @PramaanaLabs, we are designing provers that match the shape of each proof tree.
Pramaana Labs@PramaanaLabs
Math proofs are cool, but the real revolution? Verifying real-world claims in accounting, tax law, compliance, medicine & more; reliably, efficiently, and at scale. Proof trees look different across domains. One-size-fits-all won't cut it. AI prover architectures must be purpose-built to navigate the specific reasoning trees, resource constraints, and rule stability of the environment they are verifying. Read why specialized provers are the future → pramaanalabs.ai/blog/the-age-o… Authored by: @ArnavAMehta Watch this space for more to come from our Technical blog-post series.
English
Medicine is a fascinating science. Despite the enormous amount of knowledge to absorb, the best diagnosticians always have a logical method to the madness. This is an incredibly exciting first step toward formalizing that amazing clinical reasoning!
Great insights @kaush_ality and thanks for the collaboration on this @ChristineTataru!
Pramaana Labs@PramaanaLabs
Medical AI is rushing toward autonomous agents. But in the clinic, a fluent reasoning trace isn’t enough. We need proofs that are machine-checked and respects every inclusion, exclusion, and contradiction in the patient’s data. Introducing the architecture of clinical truth: formalizing diagnostic criteria in Lean 4 so every diagnosis is verifiable. No more unsafe shortcuts. Authored by: @kaush_ality @ChristineTataru Read more about it here → pramaanalabs.ai/blog/the-archi…
English
Great breakdown by @ArnavAMehta!
It's wild to think about, but proving properties about the law is actually harder and rarer than proving deep math problems. Not only is it impossible for a human to hold all the rules in mind, but the laws themselves are too ambiguous and entangled for standard solvers to brute-force. Excited to be building the infrastructure to make these massive real-world proofs possible!
Pramaana Labs@PramaanaLabs
Math proofs are cool, but the real revolution? Verifying real-world claims in accounting, tax law, compliance, medicine & more; reliably, efficiently, and at scale. Proof trees look different across domains. One-size-fits-all won't cut it. AI prover architectures must be purpose-built to navigate the specific reasoning trees, resource constraints, and rule stability of the environment they are verifying. Read why specialized provers are the future → pramaanalabs.ai/blog/the-age-o… Authored by: @ArnavAMehta Watch this space for more to come from our Technical blog-post series.
English
Thank you very much for the shoutout @svembu! This means a lot and we have been going deep and building domain-specific programming languages to capture strict nuances in areas like law and tax. Exciting times ahead!
PS: I was sitting right next to @sanjaygsub during your inspiring talk at IITM :)
Sridhar Vembu@svembu
We have been working on structural type systems and verifiable code. Best wishes to Pramaana Labs team! I am very happy to see cutting edge R&D in India. Using these approaches, I believe far more efficient as well as highly reliable AI systems are possible. That is also why I fundamentally don't agree with the trillion dollar bets being made but that is a separate topic.
English
@VarunGangal @ranjan_vittal @sanjaygsub Thanks for the kind words, @VarunGangal - we operate on the thesis that finding the right precise question is harder and more important than attempting to answer eagerly, hence our deep focus on formalization!
English
Huge congrats to @ranjan_vittal , @krishnan_rag and @sanjaygsub on emerging out of stealth with a bang!
It's been a genuine treat to've known and seen the trio in concerted action first in person and then afar past 12 years, right from office hour discussions for CS electives such as Distributed Algos and RL during our @iitmcse days, through their shared Google days and all the way down to early threads w.r.t @PramaanaLabs sharing their nuggets on miniF2F benchmarking, to their recently putting together and co-hosting the @fv_summit in SF.
Having worked on hallucinations both in production systems I contributed to as well as in my research, these as well as related misbehaviors such as sycophancy etc are a major sore to use of LLMs in many domains since they violate the implicit contracts and conventions b/w actors exchanging artifacts, be they agents or users, that the domain heavily relies on as a primitives for its existence.
Yet the basic machinery to even be able to define and detect them clearly is oft missing outside settings with well-defined dynamics such as code, math & games.
Extending Lean and adjacent means of auto-formalization for these domains, as well as then co-generating such means with outputs, i.e. "both the work and the proof of work" during inference are both pressing sub-threads to tackle and its heartening to see these being taken up by them.
The trio's attention to detail and fine taste clearly shines through even in harmoniously apt choice of name, Pramaana , the canonical term for means of evidence in Indic philosophy and epistemics!!
ranjan_raj@ranjan_vittal
Today, I'm thrilled to announce Pramaana's $27M seed, led by @khoslaventures. The foundational domains that hold the world together: tax, law, finance, healthcare; all run on certainty. Probabilistic AI can't give them that. We’ve been asked to accept wrong answers with AI as ‘hallucinations’, while in traditional software terms, it’s just a bug. And a wrong answer in such mission-critical domains is more than just a bug, it's a liability that could have catastrophic impact. We built Pramaana to deliver a 100% trustable experience to the domains that run on certainty: AI that is provably correct, not probabilistically correct. We turn statute and regulation into machine-verifiable code, so every output ships with mathematical proof of correctness. Our mission is to make AI take ownership of it’s work. Pramaana in Sanskrit stands for “means of valid knowledge”, and we’re going to achieve that by formalizing the world’s knowledge.
English
English
Heartfelt congratulations to my dear friends @ranjan_vittal, @krishnan_rag, and @sanjaygsub on the remarkable work they are doing on the formalization and verification of large language model outputs.
Having known all three since our undergraduate days, I can say with complete conviction that if anyone in the world is positioned to solve this important problem, it is them. Few people bring such relentless focus and genuine passion to every challenge they take on.
I have no doubt that @PramaanaLabs will succeed in this endeavour — and in doing so, help take humanity to new heights.
ranjan_raj@ranjan_vittal
Today, I'm thrilled to announce Pramaana's $27M seed, led by @khoslaventures. The foundational domains that hold the world together: tax, law, finance, healthcare; all run on certainty. Probabilistic AI can't give them that. We’ve been asked to accept wrong answers with AI as ‘hallucinations’, while in traditional software terms, it’s just a bug. And a wrong answer in such mission-critical domains is more than just a bug, it's a liability that could have catastrophic impact. We built Pramaana to deliver a 100% trustable experience to the domains that run on certainty: AI that is provably correct, not probabilistically correct. We turn statute and regulation into machine-verifiable code, so every output ships with mathematical proof of correctness. Our mission is to make AI take ownership of it’s work. Pramaana in Sanskrit stands for “means of valid knowledge”, and we’re going to achieve that by formalizing the world’s knowledge.
English
@ranjan_vittal is one of the most intensely driven, agentic, and helpful people out there. He leaves zero stones unturned for Pramaana Labs—gladly trading personal comfort to grind through the most grueling, unglamorous work if it gives us even a 1% edge. His helpfulness makes people genuinely root for his success.
He is undeniably brilliant and solves problems faster than anyone I know. I’ve seen this firsthand on multiple occasions, whether cracking math problems at IMOTC or making moves in our blind chess games!
Super-pumped to be building Pramaana Labs with him! 🚀
English
@TechCrunch @vkhosla @Accel @boldcapfund @NexusVP @PremjiInvest @pushmeet @SriramRajamani And we're just getting started!
If this is a problem you're passionate about solving, come build with us!
Say 'no' to hallucinations!
pramaanalabs.ai
English
@sanjaygsub and I have been friends for more than half our lives. :)
Sanjay is calm, optimistic, and relentlessly diligent. He has a quiet conviction that tunes out the industry noise / hype and delivers results - exactly what he did when helping take Gemini from a late entrant to a formidable leader in the AI space.
Now, he’s bringing that exact same focus to our shared mission: building the next wave of foundational LLMs to reason with the absolute rigor of top researchers. It’s time for AI that is actually right, not just sounding right.
Sanjay Ganapathy@sanjaygsub
Nine years at Google, three on Gemini at DeepMind, working on post-training with people I'd follow anywhere. Close enough to scale to see how far it goes, and where it stops. Today's AI is remarkable, but jagged: it hallucinates fluently. In high-stakes domains like healthcare, legal, or finance, a confident mistake is worthless. Right as I was wrestling with this, @krishnan_rag , @ranjan_vittal and I converged on a wild idea: AI that proves its own work. Proofs, not disclaimers. So I left. Nine months later, at Pramaana Labs, we're building it: a cracked team training foundation models to formalize human knowledge on a scale hitherto undreamt of, and bring formally verified AI to the real world. A future where AI is provably correct, not just probably.
English
Great to partner with @PerplexityFund! They believed in us from our earliest days, right as the foundational idea for Pramaana was first taking shape. Thrilled to have them on this journey with us!
Perplexity Fund@PerplexityFund
Not many launch videos are this inspiring-- feeling like the future is getting brighter thanks to @ranjan_vittal and team's formal AI verification for real-world domains. Congrats to @PramaanaLabs on the seed raise! Proud to back such a visionary team.
English
Can't stress enough the incredible backing and support @sathyanellore and his team, @vansht567 & @siddharth_ram, have given us. They went from being our earliest VC and acting as a sounding board from day one, to going out of their way to evangelize our mission and excite candidates. They even helped organize our recent verification summit! A truly amazing fund anyone would be lucky to be backed by.
BoldCap AI Fund@boldcapfund
Announcing our investment into @PramaanaLabs, an AI Lab for Formal Verification to formalise real world domains like tax, legal, healthcare & other mission critical domains where truth matters Congrats @ranjan_vittal @krishnan_rag @sanjaygsub Let’s Build On Large Dreams
English
@b_jishnu has been one of our earliest backers, helping us refine our vision, make critical connections, and plan from day one! It is truly amazing to be working with @b_jishnu , @746watt, and the entire @NexusVP team!
Nexus Venture Partners@NexusVP
Nexus is proud to back @PramaanaLabs in their $27M seed round! Congratulations @ranjan_vittal, @sanjaygsub, Krishnan Raghavan & team! @b_jishnu | @746watt
English
@KanuGulati has been an amazing partner from day 1! From helping with critical hiring and connections to providing thoughtful guidance across our operations, her support has been invaluable. We are so lucky to have @KanuGulati and @khoslaventures backing us!
Kanu Gulati ✈️ ICML@KanuGulati
We are thrilled to back @PramaanaLabs! As AI takes on more important work, trust can’t depend on a human checking every answer. Pramaana is building the capability to make AI outputs verifiable by design; this is a BIG step toward safer, more capable, and more useful AI. And with the right leadership in @ranjan_vittal, @krishnan_rag, @sanjaygsub and team @PramaanaLabs cc @khoslaventures @vkhosla
English
100% - couldn't agree more! Grateful he showed up and shared his unique and deeply insightful thoughts on this space - the event received immensely positive feedback!
ranjan_raj@ranjan_vittal
Last week @vkhosla anchored our summit. Injured, when stepping back would've been expected, he showed up and turned it into an example of how verification can transform healthcare. The conviction he brings to every founder, no matter how small the company, is humbling. The GOAT.
English
Thank you, @vkhosla! We are incredibly lucky to have you on board. Your early encouragement and clarity of thought have been truly inspirational for us @PramaanaLabs. We are excited to pioneer this new area of auto formalization alongside @khoslaventures!
Vinod Khosla@vkhosla
Auto formalization will be an important new area.
English
English
Love this! @krishnan_rag and I worked together in @Google and I’ve observed first hand his obsession towards solving problems. He would literally have dreams about solving user trust, and I knew back then he was built different.
Krishnan Raghavan@krishnan_rag
English