Log inSign up
Thomas Ahle
Normal Computing 🧠🌡️
6,233 posts
Image
user avatar
Thomas Ahle
Normal Computing 🧠🌡️
@thomasahle
Head of AI @NormalComputing - building OpenClaw for EDA. Ex @Meta, @BARCdk, SupWiz, @OxfordQuantum. Tweets on Math, AI, #dspy, Probability, ML. Also tensors.
Copenhagen
thomasahle.com
Joined September 2010
812
Following
8,921
Followers
  • user avatar
    Thomas Ahle
    Normal Computing 🧠🌡️
    @thomasahle
    Sep 8, 2024
    In thermodynamics, heat flows from the hotter block to the cooler one until both blocks reach the same temperature. But what if you want to transfer all the heat from block A to block B, not just equalize temperatures? Just chop up the blocks like in this animation:
    Image
    00:00
    391K
  • user avatar
    Thomas Ahle
    Normal Computing 🧠🌡️
    @thomasahle
    Jun 19, 2024
    Sam Altman: Ilya is leaving to work on a small project that's personally meaningful to him. Ilya: Single Shot Superintelligence
    user avatar
    Ilya Sutskever
    @ilyasut
    Jun 19, 2024
    I am starting a new company:
    121K
  • user avatar
    Thomas Ahle
    Normal Computing 🧠🌡️
    @thomasahle
    May 6, 2025
    Deepmind won the moment LLMs became about RL.
    user avatar
    Oriol Vinyals
    Google DeepMind
    @OriolVinyalsML
    May 6, 2025
    Ahead of I/O, we’re releasing an updated Gemini 2.5 Pro! It’s now #1 on WebDevArena leaderboard, breaking the 1400 ELO barrier! 🥇 Our most advanced coding model yet, with stronger performance on code transformation & editing. Excited to build drastic agents on top of this!
    Image
    108K
  • user avatar
    Thomas Ahle
    Normal Computing 🧠🌡️
    @thomasahle
    Jun 1, 2024
    KANs (NNs with learned functions on the edges) have a quite elegant representation using Tensor Diagrams. This chart of MLP layers also shows some neat relationship between things like ReGLUs and MoEs.
    Image
    112K
  • user avatar
    Thomas Ahle
    Normal Computing 🧠🌡️
    @thomasahle
    Nov 9, 2022
    I've been laid off from Meta. Our entire research/infra org "Probability" was cut. I deeply appreciate the people who helped me get there, and the great people I worked with for a year and half. I hope to stay in the Bay Area a while longer, if anyone needs some algorithms.
  • user avatar
    Thomas Ahle
    Normal Computing 🧠🌡️
    @thomasahle
    Jun 11, 2023
    Jax has a different definition of the identity matrix than I'm used to...
    Image
    269K
  • user avatar
    Thomas Ahle
    Normal Computing 🧠🌡️
    @thomasahle
    May 18, 2024
    Doing Matrix Calculus can be messy, specially when we need higher order derivatives. Writing them out using Tensor Diagrams makes even the Hessian Chain Rule relatively simple:
    Image
    102K
  • user avatar
    Thomas Ahle
    Normal Computing 🧠🌡️
    @thomasahle
    May 25, 2024
    I always found the tensor notation in Fast Matrix Multiplication algorithms confusing. But using tensor diagrams it's pretty easy to see what's going on:
    Image
    91K
  • user avatar
    Thomas Ahle
    Normal Computing 🧠🌡️
    @thomasahle
    Oct 19, 2023
    It is a common misconception that LLMs are just trained to "predict the next token". No. They are trained to predict an entire context window's worth of tokens, like 4k+. The gradients go end to end and the model is allowed to plan what it will say next.
    229K
  • user avatar
    Thomas Ahle
    Normal Computing 🧠🌡️
    @thomasahle
    Sep 9, 2024
    Each interaction increases entropy, but only a tiny amount—a nearly thermodynamic perfect loop.
    Image
    00:00
    77K
  • user avatar
    Thomas Ahle
    Normal Computing 🧠🌡️
    @thomasahle
    Nov 4, 2024
    The Matrix Cookbook contains the following formula, but it contains a flaw. Let's use Penrose's tensor diagrams to derive it from scratch, and correct the mistake!
    Image
    180K
  • user avatar
    Thomas Ahle
    Normal Computing 🧠🌡️
    @thomasahle
    Jul 20, 2025
    to all the people saying OpenAI's math proofs are in a "weird terse language" - might remember this by @karpathy
    user avatar
    Andrej Karpathy
    @karpathy
    Sep 16, 2024
    You can tell the RL is done properly when the models cease to speak English in their chain of thought
    68K
  • user avatar
    Thomas Ahle
    Normal Computing 🧠🌡️
    @thomasahle
    Nov 20, 2023
    Lovely table of ways to compute e^A, the exponential of a matrix, by @nhigham
    Image
    76K
  • user avatar
    Thomas Ahle
    Normal Computing 🧠🌡️
    @thomasahle
    Sep 3, 2024
    I love retro websites—so I tried to design the Tensor Cookbook (tensorcookbook.com) based on the original Matrix Cookbook aesthetics from 2006. I think it works pretty well down to the purple text, Verdana font, and everything.
    Image
    Image
    53K

New to X?

Sign up now to get your own personalized timeline!

Create account

By signing up, you agree to the Terms of Service and Privacy Policy, including Cookie Use.

Terms·Privacy·Cookies·Accessibility·Ads Info·© 2026 X Corp.
Don't miss what's happening
People on X are the first to know.
Log inSign up
Advertisement
Advertisement