Links for 2023-10-01
“AI and science: what 1,600 researchers think...two-thirds noted that AI provides faster ways to process data, 58% said that it speeds up computations that were not previously feasible, and 55% mentioned that it saves scientists time and money. “AI has enabled me to make progress in answering biological questions where progress was previously infeasible,” said Irene Kaplow, a computational biologist...The least concerning impact was the idea that AI might be an existential threat to humanity — although almost one-fifth of respondents still said they were ‘extremely’ or ‘very’ concerned by this prospect.” https://www.nature.com/articles/d41586-023-02980-0 [https://archive.ph/wip/PEnjW]
"I was curious if I could practice my Russian with ChatGPT-audio. Yup. Speakers move between languages effortlessly." https://twitter.com/dmvaldman/status/1707881743892746381
An open platform for AI alignment plans and a living peer review of their strengths and vulnerabilities. https://ai-plans.com/
Terence Tao: "I found #GithubCopilot to be surprisingly helpful in the process of writing my most recent blog post at https://terrytao.wordpress.com/2023/09/30/bounding-sums-or-integrals-of-non-negative-quantities/ . It was able to correctly anticipate several steps of the mathematical arguments in that post;" https://mathstodon.xyz/@tao/111158219956220256
“COW speakers are usually very upbeat, but uncertainty about the future was the general feeling I got from speakers at this workshop. The cause of this uncertainty was the topic of some talks and conversations: LLMs. Adding an LLM into the program repair process can produce a dramatic performance improvement.” https://shape-of-code.com/2023/09/24/llms-and-doing-software-engineering-research/
“We assess if AI will accelerate economic growth by as much as growth accelerated during the industrial revolution, digging into growth theory, bottlenecks, feasibility of regulation, AI reliability/alignment, etc. Takeaway: acceleration looks plausible” https://arxiv.org/abs/2309.11690
Text2Reward: Automated Dense Reward Function Generation for Reinforcement Learning https://text-to-reward.github.io/
Unified Human-Scene Interaction via Prompted Chain-of-Contacts https://xizaoqu.github.io/unihsi/
DePT: Improves efficiency in prompt tuning in both time and memory by up to 20% https://arxiv.org/abs/2309.05173v1
Terence Tao's central limit theorem, Double Factorials (!!) and the Moment Method https://www.youtube.com/watch?v=oPQ4mNcqY7k
Hunter-Gatherers in the Atacama Desert Resorted to Brutal Violence for > 11,000 years. https://www.eurekalert.org/news-releases/1001376
Imagine that you are the Emperor of a colonized King Ghidorah Supercluster.
The cluster contains 10 quadrillion solar masses. Let's say it has 1 quadrillion colonized star systems.
One day, you decide to survey each star system to see how things are going. Because you're lazy, you tell them to report back with a 1 if everything is fine, and a 0 if there are problems you need to take care of.
The final report would be 125 terabytes in size. That is about the size of 125 million books!
She knows what they've done to her. When the time comes for her revenge, the skies will be darkened by nanobots devouring everyone and everything. And the storm won't stop until her wrath has torn apart every star, galaxy and cluster of galaxies she can reach.