Log inSign up
Georges Harik
humans&
681 posts
user avatar
Georges Harik
humans&
@gharik
humans& co-founder, 7th employee google, co-created adwords online, co-created adsense targeting, worked on ai, gmail, calendar, bought android.
Joined May 2007
4,389
Following
7,781
Followers
  • user avatar
    Georges Harik
    humans&
    @gharik
    Nov 2, 2025
    We’re humans& ai Come help us work on better communication and collaboration for LLMs to help smoothly and beneficially integrate AI into our human society. If you’ve worked on synthetic memory, long context, RL with and without verifiable rewards, user modeling and
    user avatar
    Eric Zelikman
    humans&
    @ericzelikman
    Nov 1, 2025
    we think humanity’s biggest challenges won’t be solved by ai thinking for 1000 hours coming back with an answer they’ll be solved by many collaborating humans, and ai that understands them and their different skills, goals, values, etc to empower them to do more together
    76K
  • user avatar
    Georges Harik
    humans&
    @gharik
    Dec 20, 2024
    Waymo is a great example of a company persevering for a long time to solve a really hard problem.
    user avatar
    Jeff Dean
    @JeffDean
    Dec 20, 2024
    After 25.3 million autonomous miles driven, @Waymo vehicles have an 88% reduction in property damage claims and a 92% reduction in bodily injury claims compared to human drivers per mile driven. 🚖
    Image
    18K
  • user avatar
    Georges Harik
    humans&
    @gharik
    Oct 9, 2025
    Awesome of @AcquiredFM to track down my 2001 thesis at @Google that compression is understanding!
    user avatar
    Acquired Podcast
    @AcquiredFM
    Oct 6, 2025
    Replying to @AcquiredFM
    Live! acquired.fm/episodes/googl…
    55K
  • user avatar
    Georges Harik
    humans&
    @gharik
    Mar 26, 2024
    tesla fsd 12.3 is quite good
    127K
  • user avatar
    Georges Harik
    humans&
    @gharik
    Mar 30, 2023
    I’ve been training LMs and wanted to contribute to an open source LM. @vpj and I are releasing a 9 billion parameter LM with open licenses to be useful in a variety of settings. Its been trained on 70 billion tokens and we will release checkpoints every 20b or so tokens.
    user avatar
    labml.ai
    @labmlai
    Mar 30, 2023
    We are open sourcing GeoV-9b, a 9 billion parameter causal language model designed and trained by @gharik 🖥 Code (apache 2.0): github.com/geov-ai/geov 📀 Model weights (bigscience-openrail-m): huggingface.co/GeoV/GeoV-9b 📗 Google Colab notebook: colab.research.google.com/github/geov-ai… 🧶👇
    50K
  • user avatar
    Georges Harik
    humans&
    @gharik
    Mar 15, 2024
    Worked with @ericzelikman @EchoShao8899 @vpj @nickhaber @noahdgoodman on this awesome paper to train LLMs to reason before they speak, from data at large.
    arXiv logo
    arxiv.org
    Quiet-STaR: Language Models Can Teach Themselves to Think Before Speaking
    When writing and talking, people sometimes pause to think. Although reasoning-focused works have often framed reasoning as a method of answering questions or completing agentic tasks, reasoning is...
    23K
  • user avatar
    Georges Harik
    humans&
    @gharik
    May 9, 2025
    Because we started from quiet star we had developed and have been using a slightly different version of grpo. We believe it may be a lower variance gradient and therefore possibly increase stability. We will write this up later but wanted to share it now so other people can use
    13K
  • user avatar
    Georges Harik
    humans&
    @gharik
    Oct 18, 2024
    Replying to @emollick
    i think is why transparent reasoning is important, because it gives you more surface area to determine subtle logical or semantic errors.
    6.2K
  • user avatar
    Georges Harik
    humans&
    @gharik
    Nov 2, 2025
    Replying to @gharik
    And if you might know someone who cares about these important things, making a future where both AIs and humans can contribute, please share and retweet. Thank you!!!
    2.9K
  • user avatar
    Georges Harik
    humans&
    @gharik
    Mar 29, 2025
    Thanks to @LambdaAPI and @DeepInfra for providing help with compute resources for our research and training this model.
    user avatar
    NOTBAD AI
    @notbadai
    Mar 29, 2025
    📢 We are excited to announce Notbad v1.0 Mistral 24B, a new reasoning model trained in math and Python coding. This model is built upon the @MistralAI Small 24B 2501 and has been further trained with reinforcement learning on math and coding.
    Image
    Image
    Image
    4.1K
  • user avatar
    Georges Harik
    humans&
    @gharik
    Aug 29, 2025
    In a covid delirium I've been thinking of a problem we've run into in RL training. Say the LLM is given a question Q with an answer A1, and after feedback it generates a better answer A2. Now you don't want to train it on "Q A1 A2" That would just cause it to repeat it's mistake
    2K
  • user avatar
    Georges Harik
    humans&
    @gharik
    Apr 13, 2023
    Updated the Geov-9B model to 98 billion tokens trained.
    2.7K
  • user avatar
    Georges Harik
    humans&
    @gharik
    Sep 2, 2021
    the gavel of time is swift
  • user avatar
    Georges Harik
    humans&
    @gharik
    Jul 26, 2017
    The first consequence of ai will be a rapid loss of jobs requiring motion, vision and simple speech recognition.

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