Mo Tiwari
64 posts
@mo_tiwari
OpenAI / Stanford CS PhD / Facebook / Expanse (acquired) / Caltech
Today we open source Nomos 1. At just 30B parameters, it scores 87/120 on this year’s Putnam, one of the world’s most prestigious math competitions. This score would rank #2/3988 in 2024 and marks our first step with @hillclimbai towards creating a SOTA AI mathematician.
We flew from South Korea to host Ralphthon SF — a hackathon where you set up your AI agent, step away, and only the agent codes. Touch your laptop? Lobster costume first. 🦞 Sign-ups blew past capacity. We need a venue for 100 builders on March 28. What we built from Korea: - OpenAI sponsorship ($18K+ in prizes) - @romainhuet (Head of DevEx @OpenAI) speaking - @mo_tiwari (Google DeepMind) speaking - oh-my-opencode maintainer @q_yeon_gyu_kim (40K ⭐) judging - 8 judges including YC founders - Simultaneous event in Seoul — 200 builders across two cities If you know a space in SF Bay Area for ~100 people, please reach out. luma.com/kxoq82yq
What if we could mathematically predict how a neural network evolves during training? We developed the first mathematical framework that explains why trained networks develop the distinctive "bulk+tail" weight structure that predicts generalization, validated across transformers, vision transformers, and MLPs. 📄 arxiv.org/abs/2507.12709 1/7
Excited to release @gepa_ai's optimize_anything: a universal API for optimizing any text parameter. It consistently matches or outperforms domain-specific tools optimizing code, prompts, agent harnesses, cloud policies, even visuals! If you can measure it, you can optimize it.
What if you could mathematically guarantee your LLM won't be jailbroken? We introduce the first probabilistic certification framework for jailbreak defense that's grounded in empirically observed attack behavior, with formal proofs and validated across multiple attack types. 📄 arxiv.org/abs/2511.18721 💻 github.com/Adarsh321123/t… 1/7
What if being polite was all it took to break an LLM? We built the first fully automated pipeline for generating large-scale, psychologically-grounded multi-turn jailbreak attacks, and tested 7 models from 3 major LLM families. 📄 arxiv.org/abs/2511.19517 1/6
What if your benchmark scores are lying to you? Today, I'm excited to share @Microsoft's DevBench, the first telemetry-grounded code generation benchmark, covering six languages, and the first to combine synthetic generation with manual expert review for contamination resistance. 📄 arxiv.org/abs/2601.11895 💻 github.com/microsoft/devb… 1/7
Data centers dominate AI, but they're hitting physical limits. What if the future of AI isn't just bigger data centers, but local intelligence in our hands? The viability of local AI depends on intelligence efficiency. To measure this, we propose intelligence per watt (IPW): intelligence delivered (capabilities) per unit of power consumed (efficiency). Today’s Local LMs already handle 88.7% of single-turn chat and reasoning queries, with local IPW improving 5.3× in 2 years—driven by better models (3.2×) and better accelerators (1.7×). As local IPW improves, a meaningful fraction of workloads can shift from centralized infrastructure to local compute, with IPW serving as the critical metric for tracking this transition. (1/N)
Still buzzing from the incredible #AgenticAI Summit at @UCBerkeley on 8/2 — 2,000+ joined in person, 30,000+ tuned in online. ⚡🌍 The energy was electric—visionaries, builders & researchers shaping the future of agentic AI! Missed it? Watch the recordings: rdi.berkeley.edu/events/agentic… Huge thanks to our speakers, sponsors, @BerkeleyRDI staff, volunteers, & everyone who joined! This momentum builds on our 25K+ LLM Agents MOOC community—and we’re launching a new edition, Agentic AI, in the series this Sep! Sign up here: agenticai-learning.org/f25
Remember that Al HIlal vs Man City upset?! Come join The Beautiful Game to visualize any action on your favorite teams / players with our new zonal representation 🔥⚽️ Join us: thebeautifulgame.global 🧵1/2
