Shantanu Sharma

In 2000, he was All India Rank 1 in IIT-JEE. He took Computer Science at IIT Kanpur and finished the degree. Then he left computing altogether and today he studies the brains of mosquitoes. His name is Nitin Gupta. He graduated in 2004 with a BTech in computer science and an interest in biology that had been quietly growing through his undergraduate years. He went to the University of California, San Diego for a PhD in bioinformatics and systems biology, finishing in 2009. Then he spent a few months in a cognitive neuroscience lab there, using magnetic stimulation to study how urges for food and money leak into the motor system before a person actually moves. From 2010 to 2014 he was at the National Institutes of Health, in a laboratory studying sensory coding, learning electrophysiology. Recording the electrical conversation between neurons, one spike at a time. In July 2014, he came back to IIT Kanpur as faculty in Biological Sciences and Bioengineering. His lab asks a question that sounds simple and is not. How does a brain decide that something is worth moving towards, or running away from, without ever having learned it? They use smell, and they use insects. Grasshoppers, flies and mosquitoes. A mosquito finds you largely by carbon dioxide, picked up by sensory neurons on its maxillary palps and carried into a region of its brain called the antennal lobe. Gupta's lab works on how those circuits weigh a human body against a repellent, and how that computation becomes a decision to approach or turn away. His group also built TreadWill, a free, fully automated cognitive behavioural therapy programme for depression, designed for people in countries where therapists are scarce. In a randomised trial with 598 participants in India, users who completed at least half the modules showed significant reductions in depression and anxiety symptoms. Of the twenty students who topped IIT-JEE between 2000 and 2019, twelve now work outside India. He is not one of them. "I always knew I had to return. Movies like Swades inspired me."

I’ve always believed the No.1 application of AI should be to improve human health. That work started with AlphaFold, and now at @IsomorphicLabs with the mission to reimagine drug discovery and one day solve all disease! We are turbocharging that goal with $2.1B in new funding.

Reinforcement learning leads to better AI scientist agents! 🚀 By training models end-to-end with multi-turn RL, we’re seeing breakthroughs in reasoning and problem-solving for real biomedical research. Excited to introduce Biomni-R0 — an agentic LLM trained with this approach. On 10 real research tasks, it nearly doubles performance over its open-source base model and even surpasses closed-source frontier models by >10%. A scalable path to expert-level AI in biomedicine. Led by @RyanLi0802 @KexinHuang5 @ProjectBiomni with exciting collaboration with the SkyRL team @shiyi_c98 @NovaSkyAI. Learn more: biomni.stanford.edu/blog/bio… — open sourcing soon!

There's a new paper circulating looking in detail at LMArena leaderboard: "The Leaderboard Illusion" arxiv.org/abs/2504.20879 I first became a bit suspicious when at one point a while back, a Gemini model scored #1 way above the second best, but when I tried to switch for a few days it was worse than what I was used to. Conversely as an example, around the same time Claude 3.5 was a top tier model in my personal use but it ranked very low on the arena. I heard similar sentiments both online and in person. And there were a number of other relatively random models, often suspiciously small, with little to no real-world knowledge as far as I know, yet they ranked quite high too. "When the data and the anecdotes disagree, the anecdotes are usually right." (Jeff Bezos on a recent pod, though I share the same experience personally). I think these teams have placed different amount of internal focus and decision making around LM Arena scores specifically. And unfortunately they are not getting better models overall but better LM Arena models, whatever that is. Possibly something with a lot of nested lists, bullet points and emoji. It's quite likely that LM Arena (and LLM providers) can continue to iterate and improve within this paradigm, but in addition I also have a new candidate in mind to potentially join the ranks of "top tier eval". It is the @openrouter LLM rankings: openrouter.ai/rankings Basically, OpenRouter allows people/companies to quickly switch APIs between LLM providers. All of them have real use cases (not toy problems or puzzles), they have their own private evals, and all of them have an incentive to get their choices right, so by choosing one LLM over another they are directly voting for some combo of capability+cost. I don't think OpenRouter is there just yet in both the quantity and diversity of use, but something of this kind I think has great potential to grow into a very nice, very difficult to game eval.