Links for 2023-09-26
Amazon will invest up to $4 billion in Anthropic. The agreement is part of a broader collaboration to develop reliable and high-performing foundation models. https://www.anthropic.com/index/anthropic-amazon
OthelloGPT: How a GPT model trained to predict legal moves in the game of Othello ends up building internal representations of the board state (as shown by training a simple classifier on top the NN weights). https://arxiv.org/abs/2210.13382
Chain-of-Verification Reduces Hallucination in Large Language Models https://arxiv.org/abs/2309.11495
AI Accelerates Ability to Program Biology Like Software https://www.wsj.com/articles/ai-accelerates-ability-to-program-biology-like-software-9962a975 [https://archive.ph/Efmsz]
“Let's reverse engineer the phenomenal Tesla Optimus. No insider info, just my own analysis.” https://twitter.com/DrJimFan/status/1705982525825503282
“I made a model which takes into account AI timelines, the probability of AI going wrong, and probabilities of dying from other causes. I got that the main “end states” for my life are either dying from AGI due to a lack of AI safety (at 35%), or surviving AGI and living to see aging solved (at 43%).” https://www.lesswrong.com/posts/KFWZg6EbCuisGcJAo/immortality-or-death-by-agi-1
"ChatGPT can now see, hear, and speak. Rolling out over next two weeks, Plus users will be able to have voice conversations with ChatGPT (iOS & Android) and to include images in conversations (all platforms)." https://openai.com/blog/chatgpt-can-now-see-hear-and-speak
“Speak in your voice in the listener’s language, not your own, using the latest @OpenAI voice synthesis technology.” https://twitter.com/gdb/status/1706381547526558002
NeuroMechFly 2.0, a framework for simulating embodied sensorimotor control in adult Drosophila https://www.biorxiv.org/content/10.1101/2023.09.18.556649v1
Brain GO - Electric Wheelchair Controlled by Brain Waves https://www.youtube.com/watch?v=Br2d2T3ZTzY
Biomarkers from brain data can predict depression relapse 5 weeks BEFORE a patient self-reports. https://www.nature.com/articles/s41586-023-06541-3
With only 160 short gene sequences, race can be determined with nearly 100% accuracy for Europeans, Asians, and Africans. https://www.cell.com/ajhg/fulltext/S0002-9297(07)60574-6
"The CEA will stop at nothing to ensure the survival of human civilization" said CEA Director Yudkowsky. https://twitter.com/PatrickJBlum/status/1706354581989626323 (There is a short SF series about this: Bayeswatch https://www.lesswrong.com/s/TjdhvTSptCYakw3Lc)
Some crazy but fun rumors: https://twitter.com/Simeon_Cps/status/1706078819617063304
Given these rumors, it's again worth considering what this would mean in practice.
People who compare artificial intelligence to Web 3.0 or Bitcoin have not understood what it is all about. Intelligence is the most powerful force in the universe. Artificial General Intelligence (AGI) is literally the last invention that humans need to make because all other inventions will follow automatically.
Think about what it means to be able to copy general intelligence: labor becomes a free commodity like sunlight. That is, it can be consumed in as much quantity as needed without diminishing its availability.
AGI is the seed that grows into a forest all by itself because the robots will build themselves. AGI is the automation of automation. It will multiply its existing intelligence by recursively improving itself.
If AGI can be aligned with human values, we will end up in a world where anyone can be far more prosperous than the richest person alive today, regardless of their own abilities. Even cosmic engineering projects will become feasible using exponentially self-replicating von Neumann probes.
India's total fertility rate (TFR) - children per woman - fell to 2 in the latest assessment period, for 2019-2021, from 3.4 in 1992-93, according to a government report issued in October. It estimated that the average must be 2.1 for the population to reproduce itself.