Log inSign up
Image
user avatar
Andreas Stuhlmüller
@stuhlmueller
CEO @elicitorg | It's all wrong and early
stuhlmueller.org
Joined July 2008
196
Following
3,280
Followers
  • Pinned
    user avatar
    Andreas Stuhlmüller
    @stuhlmueller
    Jun 20
    This time with Nathan we zoomed in on explicit world models as evolving representations that answer prediction/intervention/counterfactual questions. Continual learning outside the weights. IMO still a lot of alpha here
    user avatar
    The Cognitive Revolution Podcast
    @CogRev_Podcast
    Jun 18
    🧠NEW EPISODE: @stuhlmueller and @jungofthewon , co-founders of @elicitorg , are back on The Cognitive Revolution with @labenz after two years — and the mission hasn't moved an inch: radically improve the quality of reasoning behind high-stakes decisions. What's changed is how
    Image
    00:00
    928
  • user avatar
    Andreas Stuhlmüller
    @stuhlmueller
    Sep 20, 2022
    Language model papers at NeurIPS 2022 that sound interesting to me and that I hadn't seen before (thread)
  • user avatar
    Andreas Stuhlmüller
    @stuhlmueller
    Oct 17, 2020
    Language model experiment @oughtinc: Take a vague forecasting question ("What will the future of robotics look like?"), generate measurable subquestions ("How many industrial robots sold per year?"), then generate a data source for each measurement ("Intl Federation of Robotics")
    Image
    00:00
  • user avatar
    Andreas Stuhlmüller
    @stuhlmueller
    Aug 12, 2022
    I've updated the @oughtinc machine learning reading list with 30+ papers from the last 6 months: PaLM, Chinchilla, Instruct, Grokking, T-Few, Chain of Thought, Self-Consistency, Minerva, Selection-Inference, Cascades, Plex, Forecasting, etc
    Image
    docs.google.com
    Elicit Machine Learning Reading List
    The most recent version of this list now lives at: https://github.com/elicit/machine-learning-list The list below is an older copy. Elicit Machine Learning Reading List Purpose The purpose of this...
  • user avatar
    Andreas Stuhlmüller
    @stuhlmueller
    Nov 3, 2023
    Underappreciated thought/feeling by @KatjaGrace - definitely feels like people haven't grappled with the consequences of AI as much as seems right, even the people who worry about the consequences a lot worldspiritsockpuppet.com/2023/11/03/the…
    Image
    23K
  • user avatar
    Andreas Stuhlmüller
    @stuhlmueller
    Aug 22, 2023
    Thread of reasons to work with us on Elicit (1/♾)
    24K
  • user avatar
    Andreas Stuhlmüller
    @stuhlmueller
    May 17, 2021
    language models + dataframes = ❤️
    Image
    00:00
  • user avatar
    Andreas Stuhlmüller
    @stuhlmueller
    Jul 26, 2021
    1. select text in browser, pdf reader, anywhere else 2. press command-option-enter 3. see hierarchical outline a la @RoamResearch @WorkFlowy
    Image
    00:00
  • user avatar
    Andreas Stuhlmüller
    @stuhlmueller
    Jun 10, 2021
    a cryptocurrency that incentivizes miners to collectively train a gpt3-like model and keep its knowledge up to date, moving training to wherever energy is cheapest
  • user avatar
    Andreas Stuhlmüller
    @stuhlmueller
    Oct 7, 2022
    New beta feature in @elicitorg: Synthesize the top papers into a summary answer. Updates when you remove irrelevant papers
    Image
    00:00
  • user avatar
    Andreas Stuhlmüller
    @stuhlmueller
    Jun 22, 2021
    my notes on management that I've so far only shared with managees at ought
    Image
    Working with Andreas - a user guide
    From stuhlmueller.org
  • user avatar
    Andreas Stuhlmüller
    @stuhlmueller
    Aug 9, 2024
    This law almost applied to me, and I have PhD friends who have to leave the US because of it. I probably wouldn't have founded Elicit in the US if I had to go back to Germany for two years after my postdoc
    user avatar
    Jeremy Neufeld
    IFP
    @JeremyLNeufeld
    Aug 9, 2024
    An obscure immigration law called 212(e) forces certain J-1 visa holders to return home for two years before they can get another visa New research from @m_clem, @AmyMNice, and me shows the rule’s cost for the first time: It makes us kick out ~40,000 high-skilled workers a year
    Image
    Image
    14K
  • user avatar
    Andreas Stuhlmüller
    @stuhlmueller
    Sep 28, 2022
    Out now: - The Interactive Composition Explorer (ICE), a Python library for writing and debugging compositional language model programs github.com/oughtinc/ice - The Factored Cognition Primer, a tutorial that shows using examples how to write such programs primer.ought.org
    Image
    00:00
  • user avatar
    Andreas Stuhlmüller
    @stuhlmueller
    Nov 1, 2021
    We finetuned a language model on answering science questions given abstracts, live on elicit.org now. Because it starts with paper search, not free generation, it usually doesn't hallucinate. Next step is indexing full PDFs
    Image
    00:00