Links for 2023-05-07
scGPT: first foundation large language model for single-cell biology. It is a transformer network trained on gene expression that can do cell-type annotation, genetic perturbation prediction, and gene network inference. https://twitter.com/simocristea/status/1654581096498229250 (paper: https://www.biorxiv.org/content/10.1101/2023.04.30.538439v1)
Predict what a mouse sees by decoding brain signals https://www.youtube.com/watch?v=bWZiPT8IhRY
Why aren’t more of us working to prevent AI hell? https://www.lesswrong.com/posts/5Jmhdun9crJGAJGyy/why-are-we-so-complacent-about-ai-hell
S-risk reduction is separate from alignment work https://forum.effectivealtruism.org/posts/8yaQ6i3oaFLprsFyb/ai-alignment-researchers-may-have-a-comparative-advantage-in#S_risk_reduction_is_separate_from_alignment_work
What does it take to catch a Chinchilla? Verifying Rules on Large-Scale Neural Network Training via Compute Monitoring — “The framework's primary goal is to provide governments high confidence that no actor uses large quantities of specialized ML chips to execute a training run in violation of agreed rules.” https://arxiv.org/abs/2303.11341
An architecture that combines deep neural networks and vector-symbolic models https://techxplore.com/news/2023-03-architecture-combines-deep-neural-networks.html
Multimodal Procedural Planning via Dual Text-Image Prompting https://arxiv.org/abs/2305.01795
Visual Chain of Thought: Bridging Logical Gaps with Multimodal Infillings https://arxiv.org/abs/2305.02317
We're Afraid Language Models Aren't Modeling Ambiguity https://arxiv.org/abs/2304.14399
“Deep-learning system explores materials’ interiors from the outside: A new method could provide detailed information about internal structures, voids, and cracks, based solely on data about exterior conditions.” https://news.mit.edu/2023/deep-learning-system-explores-interiors-0428
Multi-Party Chat: Conversational Agents in Group Settings with Humans and Models https://arxiv.org/abs/2304.13835
Contra Yudkowsky on Doom from Foom #2 https://www.lesswrong.com/posts/LF3DDZ67knxuyadbm/contra-yudkowsky-on-doom-from-foom-2
Heritability, race, and police bias: Kareem Carr's blunder https://kirkegaard.substack.com/p/heritability-race-and-police-bias
African IQ's: reality vs. Kareem Carr https://kirkegaard.substack.com/p/african-iqs-reality-vs-kareem-carr
Stupid? No. Unfamiliar? Yes. The meaning of low mean IQs in developing nations https://russellwarne.com/2023/04/29/stupid-no-unfamiliar-yes-the-meaning-of-low-mean-iqs-in-developing-nations/
“Lithuania is the world’s only country in a state of open confrontation with both Russia and China” [Wall Street Journal] https://archive.is/oiujd
This one graph is more informative than most sociology degrees.
Paper: https://www.nature.com/articles/s41539-017-0005-6
See also: Genetic confounding: When you hear "privilege" or "discrimination" always think "genetic confounding" or risk wasting billions on futile efforts https://axisofordinary.substack.com/p/genetic-confounding