user avatar
Sören Mindermann
@sorenmind
Postdoc with Yoshua Bengio, Mila
Oxford
Joined May 2016
177
Following
1,757
Followers
  • Pinned
    user avatar
    Sören Mindermann
    @sorenmind
    Dec 15, 2020
    Super excited to share that **Inferring the effectiveness of government interventions against COVID-19** was just published in Science !! science.sciencemag.org/lookup/doi/10.… Work done with amazing collaborators @JanMBrauner, @MrinankSharma ... 1/
  • user avatar
    Sören Mindermann
    @sorenmind
    Jun 16, 2022
    Tired of waiting 💤 while your model trains? Try skipping points that are already learned, not learnable or not worth learning! Robustly reduces required training steps 🏎 by >10x ! to reach the same accuracy on big web-scraped data 📜ICML 2022 paper: arxiv.org/abs/2206.07137
  • user avatar
    Sören Mindermann
    @sorenmind
    Jun 19, 2024
    Francois Chollet says the ARC challenge for AI is easy even for 5 year olds. Can a 5 year old help me with this please?
    796K
  • user avatar
    Sören Mindermann
    @sorenmind
    May 2, 2023
    Our ICML submission on the difficulty of aligning AGI got rejected despite unanimous-accept reviews (7/7/5). Read the paper ICML chairs don't want you to see 🤫:
    arXiv logo
    arxiv.org
    The Alignment Problem from a Deep Learning Perspective
    In coming years or decades, artificial general intelligence (AGI) may surpass human capabilities across many critical domains. We argue that, without substantial effort to prevent it, AGIs could...
    69K
  • user avatar
    Sören Mindermann
    @sorenmind
    Mar 27, 2021
    “Understanding the effectiveness of government interventions in Europe’s second wave of COVID-19” -- a result of a new collaboration between @Oxford, @Imperial, @FHIOxford, @OATML_Oxford, @uni_copenhagen, @LSHTM, @Cambridge_Uni, @TheCrick... 1/ ➡️Paper: medrxiv.org/content/10.110…
  • user avatar
    Sören Mindermann
    @sorenmind
    Oct 30, 2023
    Our new consensus paper on AI risks & measures, including policy and technical measures, has made the The Guardian's front page! Whereas Biden’s "sweeping" executive order from today holds no-one responsible.. 1/🧵
    46K
  • user avatar
    Sören Mindermann
    @sorenmind
    Dec 11, 2021
    Well-deserved comment on my interview with DW News
  • user avatar
    Sören Mindermann
    @sorenmind
    Jul 27, 2020
    Excited to share a major update on how effective 8 interventions have been against COVID-19 transmission in 41 countries. Paper: medrxiv.org/content/10.110…
    Note caveats on masks in the paper, they're context-dependent
  • user avatar
    Sören Mindermann
    @sorenmind
    Apr 2, 2024
    Life achievement unlocked: someone calls your paper 'seminal' on Twitter <3
    user avatar
    Agus 🔸
    @austinc3301
    Mar 27, 2024
    Replying to @Hongzhou_Luan
    If I had to think for one seminal paper, maybe I would go for The Alignment Problem from a Deep Learning Perspective by @RichardMCNgo, @justanotherlaw and @sorenmind, which is also pretty comprehensive arxiv.org/abs/2209.00626
    1.5K
  • user avatar
    Sören Mindermann
    @sorenmind
    Jun 16, 2022
    Replying to @sorenmind
    Training on big web-scraped data can take ages 💤 But lots of compute and time is wasted on redundant and noisy points that are already learned, not learnable, or not even worth learning.
  • user avatar
    Sören Mindermann
    @sorenmind
    Oct 27, 2023
    Replying to @geoffreyhinton
    ”Even at Inflection, with the compute that we have, will be 100x larger than the current frontier models in the next 18 months.“ Geoffrey‘s tweet references this quote from Inflection CEO Mustafa Suleyman
    9.7K
  • user avatar
    Sören Mindermann
    @sorenmind
    Feb 25, 2020
    Replying to @fchollet
    Clearly one is more confusing than the other...
  • user avatar
    Sören Mindermann
    @sorenmind
    Jun 16, 2022
    Replying to @sorenmind
    Work done at @OATML @CohereAI with great collaborators @JanMBrauner, @MrinankSharma @mtrazzak, @BlackHC, Winnie Xu, Ben Höltgen, @adrien_morisot, @aidangomezzz, @seb_far, @yaringal ⌨️Code:
    GitHub - OATML/RHO-Loss
    From github.com
  • user avatar
    Sören Mindermann
    @sorenmind
    Jun 16, 2022
    Replying to @sorenmind
    That’s where Reducible Holdout Loss Selection (RHO-LOSS) comes in. We select points that most reduce the generalisation loss and show that this objective has a simple and cheap but close approximation (line 7).