new paper. During learning, why does it help to practice easy examples before hard ones? We provide a theory, based on deep learning models. TL;DR - curricula help with unlearning rather than learning. arxiv.org/abs/2611.04565. Led by @RubruckJirko and Alice Zhang, with @SaxeLab
summerfieldlab @summerfieldlab.bsky.social
@summerfieldlab
Studying human behaviour and brain function, and how to build safe and beneficial technology
901 Following 8.6K Followers 819 Posts
Latest posts 5
AI research stays true for almost exactly the amount of time it takes an academic journal to publish it
How long does a finding about LLMs stay true?View quoted post on X
A new paper re-tested 50 widely cited findings about LLM behavior (biases, hallucination, sycophancy, etc).
Within 15 months, most no longer held or had substantially reduced effects. The shelf life of LLM findings is short.
Almost half (44%) of the tested claims are now expired in today’s models, 34% are weakener, 16% persisted, and 6% have strengthened relative to the original claim
The only effects that got stronger were about sycophancy, LLM ingroup bias, and creativity.
This means we should be cautious when interpreting the long term durability of research on LLMs.
zou-group.github.io/llm-claims-she…
please do submit an essay for this!
AI agents could make organizations far more productive, but leave people more isolated and less autonomous.View quoted post on X
We've teamed up with @inherent_labs on an essay prize exploring how organizations can harness AI while strengthening human autonomy and fulfillment.
We're thrilled that @lugaricano, @gleech, @divyasiddarth, and Chris Summerfield are joining us as judges.
blog.cosmos-institute.org/p/essay-contes…
thank you! and of course everyone should read @MasudHusain 's book Our Brains, Our Selves, which is illuminating and entertaining in equal measure!
Fran Ashcroft and I discuss Chris Summerfield's insightful and very stimulating book on AI -- THESE STRANGE NEW MINDS -- on her podcast A Good Science Read podcasts.ox.ac.uk/good-science-r…View quoted post on X
I have read Matt's book and you should too. It's a take that you won't find elsewhere - informed by a deep understanding of how tech works, and how the state works. recommend!
My book, AI and Political Freedom, is out today.View quoted post on X
It argues that AI safety isn’t only a technical and regulatory problem. No matter what measures we take on those fronts, AI won’t be safe unless we also reinforce the foundations of constitutional democracy. The book lays out specific steps for ensuring that political freedom survives alongside powerful AI.
Link in the first reply.