Links for 2022-01-04
"A Neural Network Solves and Generates Mathematics Problems by Program Synthesis: Calculus, Differential Equations, Linear Algebra, and More", Drori et al 2021 (Codex) https://arxiv.org/abs/2112.15594
"Player of Games", Schmid et al 2021 {DM} (generalizing AlphaZero to imperfect-information games) https://arxiv.org/abs/2112.03178
U.S. vs. China Rivalry Boosts Tech—and Tensions -> Militarized AI threatens a new arms race — “Tang Jie, the Tsinghua University professor leading the Wu Dao project, said in a recent interview that the group built an even bigger, 100 trillion-parameter model in June, though it has not trained it to “convergence,” the point at which the model stops improving. “We just wanted to prove that we have the ability to do that,” Tang said.” https://spectrum.ieee.org/china-us-militarized-ai
John Wentworth on How To Get Into Independent Research On AI Alignment. "I’m an independent researcher working on AI alignment and the theory of agency. I’m 29 years old, will make about $90k this year, and set my own research agenda. I deal with basically zero academic bullshit...best of all, I work on some really cool technical problems which I expect are central to the future of humanity. If your reaction to that is 'Where can I sign up?', then this post is for you." https://www.lesswrong.com/posts/P3Yt66Wh5g7SbkKuT/how-to-get-into-independent-research-on-alignment-agency
AI Safety Needs Great Engineers: If you think you could write a substantial pull request for a major machine learning library, then major AI safety labs want to interview you today. https://forum.effectivealtruism.org/posts/DDDyTvuZxoKStm92M/ai-safety-needs-great-engineers
The Omicron lab leak hypothesis explained https://bprice.substack.com/p/lab-leak-20
Scott Alexander’s Links for December https://astralcodexten.substack.com/p/links-for-december
The demise of Scientific American: Guest post by Ashutosh Jogalekar https://scottaaronson.blog/?p=6202
Some quotes about the climate "emergency" we are facing:
1. “Take even worst-case estimates that climate change will lower GDP by 5-10% in the year 2100. Compared to growth, that's couch change. At our current tragically low 2% per year, without even compounding (or in logs), GDP in 2100 will be 160% greater than now. Climate change will make 2100 be as terrible as... 2095 would otherwise be. If we could boost growth to 3% per year, GDP in 2100 will be 240% greater than now, an extra 80 percentage points. 8% in 80 years is one tenth of a percent per year growth. That's tiny.” [Source: https://johnhcochrane.blogspot.com/2021/07/how-much-does-climate-change-actually.html]
2. “One thing he wants to make very clear is that all the paths, even the hottest ones, show improvements in human well-being on average. IPCC scientists expect that average life expectancy will continue to rise, that poverty and hunger rates will continue to decline, and that average incomes will go up in every single plausible future, simply because they always have. “There isn’t, you know, like a Mad Max scenario among the SSPs,” O’Neill said. Climate change will ruin individual lives and kill individual people, and it may even drag down rates of improvement in human well-being, but on average, he said, “we’re generally in the climate-change field not talking about futures that are worse than today.”” [Source: https://amp.theatlantic.com/amp/article/620605/]
3. “If you live in the first world, with the exception of low lying cities like New Orleans or Venice, most people aren’t actually going to notice much of a difference.” [Source: https://www.youtube.com/watch?v=Df5fooIiasY]