Links for 2023-05-15
How worried about AI risk will we feel in the future, when we can see advanced machine intelligence up close? We should worry accordingly now. https://joecarlsmith.com/2023/05/08/predictable-updating-about-ai-risk
“Biggest prosaic-LLM-alignment breakthrough of 2023 imo: turns out that, in GPT-2-XL, activation vectors in the residual steam have the same kind of affine structure as good old word2vec, but higher layers become emotional, then conceptual, then cognitive” — davidad https://www.alignmentforum.org/posts/5spBue2z2tw4JuDCx/steering-gpt-2-xl-by-adding-an-activation-vector
A useful way to intuitively understand the transformer architecture? https://www.lesswrong.com/posts/euam65XjigaCJQkcN/an-analogy-for-understanding-transformers
MIT engineers’ new technique analyzes the 3D organization of the genome at a resolution 100 times higher than before. https://news.mit.edu/2023/unprecedented-view-3d-genome-0508
De novo evolution of macroscopic multicellularity [The New York Times] https://archive.is/1yCVn
Enhancing cognitive control in amnestic mild cognitive impairment via at-home non-invasive neuromodulation in a randomized trial https://www.nature.com/articles/s41598-023-34582-1
"free public education in the 19th century and reduction of nepotism have not increased social mobility. Nor is there any sign modern economic growth has done so. The expansion of the franchise in the 19th and 20th centuries similarly has had no effect" https://twitter.com/robkhenderson/status/1655695437938999297
“When Communists get a chance to live their dream and kill the skilled, it must bring down a nation's human capital. New paper on Cambodia compares two provinces: one had even more politicide than the other, and afterwards had even less schooling and literacy & more poverty” https://twitter.com/whyvert/status/1656116335481372674
The statistical confusion about crime rates is off the charts https://heterodoxthinking.substack.com/p/the-statistical-confusion-about-crime
Long interview with the Commander-in-Chief of the Armed Forces of Ukraine https://youtu.be/TadXxP_26V8?t=190
Soon, small talk will be automated by LLMs + voice cloning. Imagine being at a party and using your smart glasses to do more interesting things like reading a book, while an LLM fine-tuned to your personal data does all the boring socializing. All you have to do is randomly open and close your mouth as if you're actually talking.