AI Dynamics

Global AI News Aggregator

About

RESEARCH

  • Lean community contributions to proof tooling and Mathlib development

    By

    @animaanandkumar

    –

    RT @Robertljg: Honestly, a huge part of this is thanks to the incredible Lean community @leanprover ! Mathlib, proof tooling, and years of…

    → View original post on X — @animaanandkumar, 2026-03-02 00:12 UTC

  • Scaling Laws and Compute in AI Model Development

    By

    @tunguz

    –

    a lot of energy over the past few years went into trying to solve problems that would naturally resolve themselves with larger models/memory/compute.

    → View original post on X — @tunguz

  • Lean Community Infrastructure Enables LLM Advancement in Proof

    By

    @animaanandkumar

    –

    Honestly, a huge part of this is thanks to the incredible Lean community @leanprover ! Mathlib, proof tooling, and years of infrastructure work made this possible —Lean is far more usable today because of that collective effort. And yes, LLMs have gotten much better at Lean lately 🙂 definitely helps accelerate learning and prototyping. Evan Chipman (@evanchipman) This is the 3rd time this week I thought “someone ought to make x” then open this app and see a team announce x. The speed of this era is disorienting. — https://nitter.net/evanchipman/status/2028157017597374837#m

    → View original post on X — @animaanandkumar, 2026-03-01 22:57 UTC

  • TorchLean: First Fully Verified Neural Network Framework in Lean
    TorchLean: First Fully Verified Neural Network Framework in Lean

    By

    @animaanandkumar

    –

    Super excited to release TorchLean!! I’m happy to answer questions and would love to discuss verified NNs + theorem proving especially what it’ll take for the field to become widely usable in real ML systems. Blog post + codebase release soon! Prof. Anima Anandkumar (@AnimaAnandkumar) We’re excited to release TorchLean which is the first fully verified neural network framework in Lean. The Lean community has largely focused on pure mathematics. TorchLean expands this frontier toward verified neural network software and scientific computing. With the recent release of CSlib, we see this as another step toward a fully verified ML stack. We support features: 1. Executable IEEE-754 floating-point semantics (and extensible alternative FP models) verified tensor abstractions with precise shape/indexing semantics 2. Formally verified autograd system for differentiation of NN programs Proof-checked certification / verification algorithms like CROWN (robustness, bounds, etc.) 3. PyTorch-inspired modeling API with eager-style development + export/lowering to a shared IR for execution and verification Project page: leandojo.org/torchlean.html Paper: [2602.22631] TorchLean: Formalizing Neural Networks in Lean Work done @Robertljg, Jennifer Cruden, Xiangru Zhong, @huan_zhang12 and @AnimaAnandkumar. #MachineLearning #ScientificComputing #Lean — https://nitter.net/AnimaAnandkumar/status/2027907453908857298#m

    → View original post on X — @animaanandkumar, 2026-03-01 22:38 UTC

  • Top AI models stay close in a fast-moving packed peloton

    By

    @aymericroucher

    –

    Funny how the top models from different labs stay within a tiny distance from each other. The peloton is moving extremely fast, but stays quite packed, no breakaway so far. And that is despite large differences, e.g. in available compute. What's the forcing function that makes

    → View original post on X — @aymericroucher

  • Modular Agentic RAG system with LangGraph and HITL

    By

    @kirkdborne

    –

    Build a modular Agentic RAG system with LangGraph, conversation memory, and human-in-the-loop query clarification using this GitHub repo:

    → View original post on X — @kirkdborne

  • WiFi signals turned into radar to see through walls and estimate poses

    By

    @linusekenstam

    –

    We just turned WiFi signals into a radar that can see through walls and estimate exact poses of people. Surveillance just got order of magnitude more easy todo. No need for cameras. Git hub repo close to 12k https://
    github.com/ruvnet/wifi-de
    nsepose
    … https://
    x.com/BoWang87/statu
    s/2027941789848514643/video/1
    …

    → View original post on X — @linusekenstam

  • OpenAI criticism deserves nuance: frontier innovation requires visible risk

    By

    @arrakis_ai

    –

    I’m honestly tired of watching OpenAI get cast as the default villain in every AI debate. They try something bold, it’s dangerous. They move fast, it’s irresponsible. They partner, it’s corruption. They compete, it’s opportunism. Meanwhile, plenty of other companies move quietly, wait for validation, copy what works, avoid the hardest calls — and somehow escape the same scrutiny. Let’s be honest. OpenAI ships at scale. They deploy first. They test boundaries in public. That means they make visible mistakes. Visible tradeoffs. Visible bets. But that’s also what pushing a frontier looks like. If you’re the company actually attempting moonshots, integrating with institutions, scaling globally, and defining new categories, you’re going to absorb more risk and more criticism than everyone standing safely behind you. Do we really believe other AI companies aren’t navigating the same ethical gray zones? The same regulatory ambiguity? The same pressure between innovation and governance? Or is it just easier to project all systemic anxiety onto the biggest target? The standard keeps rising for OpenAI. Higher than for startups. Higher than for open source projects. Higher than for incumbents moving quietly in the background. Criticism is necessary. Accountability matters. But pretending only one company operates in tension with power, policy, and profit feels intellectually dishonest. Frontier innovation is messy. Governance is incomplete. The incentives are complex. If we want responsible AI, we should demand better systems, not just better villains. Because the reality is this: The companies actually trying to reshape infrastructure will always look riskier than the ones waiting to copy the outcome. And asking them to innovate without taking risk is asking them to do nothing at all.

    → View original post on X — @arrakis_ai, 2026-03-01 03:32 UTC

  • Formal Verification for Neural Networks in Lean

    By

    @animaanandkumar

    –

    Verification is a cornerstone for reasoning. Recent progress in mathematical reasoning and theorem proving has relied on LLMs + formal verification in Lean. But so far, Lean has focused on pure math, and lacks support for verification involving neural nets themselves. Applications include: 1. Certified robustness in neural networks, crucial in safety critical applications such as neural control. 2. Physics informed neural networks, which have been used to prove singularity problems related to the Millennium prize problem in fluid dynamics. 3. Theory related to neural networks such as approximation bounds, effect of quantization etc.

    → View original post on X — @animaanandkumar, 2026-03-01 01:04 UTC

  • Understanding Composite AI: Innovation in Machine Learning

    By

    @terence_mills

    –

    What Is Composite AI? #AI #AIio #AIInnovation #ML #DataScience #Futureofwork @HaroldSinnott @fogoros @iainljbrown @NandoDF @katecrawford @drhassanrashidi @YuHelenYu ow.ly/SMmq30sTyEB

    → View original post on X — @terence_mills, 2026-03-01 01:00 UTC