Log inSign up
Image
user avatar
hattie
@oh_that_hat
Building friendly AI.
San Francisco
Joined December 2010
890
Following
12.5K
Followers
  • user avatar
    hattie
    @oh_that_hat
    Jul 23, 2025
    Interesting piece by Matt Levine on the huge AI salaries: “I tell you what, if Meta Platforms Inc. paid me a $100 million signing bonus to come work for their artificial intelligence business, I would be the most dedicated worker they have ever seen until the check cleared!
    174K
  • user avatar
    hattie
    @oh_that_hat
    Jul 13, 2024
    I’m excited to share that I recently joined @AnthropicAI! One thing that struck me the most so far is how humble and collaborative everyone is (besides being totally ✨brilliant✨)… Not just “humble and collaborative”, but like 2 std above what I expected. Pretty cool!
    216K
  • user avatar
    hattie
    @oh_that_hat
    Oct 25, 2023
    What algorithms can Transformers learn? They can easily learn to sort lists (generalizing to longer lengths), but not to compute parity -- why? 🚨📰 In our new paper, we show that "thinking like Transformers" can tell us a lot about which tasks they generalize on!
    Image
    388K
  • user avatar
    hattie
    @oh_that_hat
    Aug 21, 2019
    Love this book dedication
    Image
  • user avatar
    hattie
    @oh_that_hat
    Mar 27, 2022
    How do people deal with seeing their parents age? I find it to be nearly unbearable
  • user avatar
    hattie
    @oh_that_hat
    Nov 17, 2022
    “LLMs can’t even do addition” 📄🚨We show that they CAN add! To teach algos to LLMs, the trick is to describe the algo in enough detail so that there is no room for misinterpretation w/ @Azade_na @hugo_larochelle @AaronCourville @bneyshabur @HanieSedghi arxiv.org/abs/2211.09066
    Image
  • user avatar
    hattie
    @oh_that_hat
    Sep 21, 2025
    It takes different skills to be 3-9 months ahead of the curve than 1-4 years ahead. To be 3-9 months ahead, you just need to be observant about the current state of things and notice gaps. This lets you do things like publishing papers. To be 1-4 years ahead requires sustained
    64K
  • user avatar
    hattie
    @oh_that_hat
    Mar 3, 2023
    Friend: I’m listening to a talk about grounding transformers for safety purposes Me: ohh like AI safety? Friend: like preventing fires Me: so “grounding transformers” means… Friend: like actual ground wires to actual electrical transformers
    58K
  • user avatar
    hattie
    @oh_that_hat
    Jul 11, 2025
    we wanted a diffusion tutorial that's gentle and fun because we banged our heads for a long time trying to learn
    user avatar
    Pramod Goyal
    @goyal__pramod
    Jul 10, 2025
    A beautiful paper that goes through Diffusion step by step, explaining the entire math of it from the beginning.
    Image
    46K
  • user avatar
    hattie
    @oh_that_hat
    Aug 14, 2025
    AI models “think” in two ways: - in the latent space over layers - in the token space over a sequence Latent space = natural talent, chain of thought = hard work. Just like for humans, hard work can get you far, but talent sets the ceiling. This is why pretraining can’t die.
    72K
  • user avatar
    hattie
    @oh_that_hat
    Nov 30, 2021
    Learning ML is easier than ever before, but applying to grad school in ML has only gotten harder, esp for people from non-traditional backgrounds. If you come from a non-stem background and are thinking about transitioning to a career in AI, I'm happy to chat/answer any Qs!
  • user avatar
    hattie
    @oh_that_hat
    Aug 6, 2025
    Top AI labs have 1000x more applicants than openings. A job there is a dream come true for many PhD students. But for those at Berkeley or Stanford, working at one of these labs is a birthright, because “where else are you gonna go?” Physical proximity still matters so much in
    64K
  • user avatar
    hattie
    @oh_that_hat
    Jun 14, 2024
    We spent months trying to learn about diffusion models, and I’ve found it hard to learn - diffusion is mathy and spans many different fields, and it’s full of small design choices that are practically important yet obscure This tutorial is what we wish we had, hope it’s helpful!
    user avatar
    Preetum Nakkiran
    @PreetumNakkiran
    Jun 14, 2024
    Our tutorial on diffusion & flows is out! We made every effort to simplify the math, while still being correct. Hope you enjoy! (Link below -- it's long but is split into 5 mostly-self-contained chapters). lots of fun working with @ArwenBradley @oh_that_hat @advani_madhu on this
    Image
    85K
  • user avatar
    hattie
    @oh_that_hat
    Feb 2, 2022
    [New paper 🚨] You’ve heard of catastrophic forgetting, but have you heard of *fortuitous* forgetting? In this paper, we show how forgetting can be a friend of learning in artificial neural nets. With @AnkitKVani, @hugo_larochelle, @AaronCourville (1/6) arxiv.org/abs/2202.00155