AI Reading for Sunday October 13
Lovingly curated by man and machine
Meta is proud of its AI slop, like a bad dog after a rampage of destruction. - The Verge
Drone swarms monitor US military installations. - WSJ
AI Pioneer Yann LeCun Thinks AI Is Dumber Than a Cat. - WSJ
"I was a beta tester for the Nobel prize-winning AlphaFold AI – it’s going to revolutionize health research" - The Conversation
Oklo begins Idaho site prep with aim to develop nuclear reactors by 2027, no nuclear permits yet but DoE let them start construction prep at their designated site. - Interesting Engineering
How ChatGPT can help with procrastination and difficult emails. - FT
AI snitch cuts down abusive trash talk in online games. - pcgamer
Using ChatGPT to improve your prompts. - Geeky Gadgets
One thing I think chat UIs could use is prompt templates with a few variables, I use Librechat , there is also AIPRM.
Try: "From all of our interactions that you stored in memory, what is one thing that you can tell me about myself that I may not know about myself?"
Using AI to recursively improve itself - arXiv.org
Write an agent to perform a task, then write an agent to think about the first agent and iteratively improve it via Q-learning or policy iteration.
Yesterday I highlighted Dario Amodei’s 15,000-word blog post about upsides of AI, if you’re interested in this stuff, it’s worth an hour or so to read.
It’s an important thought experiment, what could we do if we had infinite intelligent compute.
I don’t see Sam Altman or Elon Musk thinking this hard about this stuff, Altman seems callow, and Musk has a messianic God complex. (maybe all of them do)
I guess I understand why the Valley boys are talking about spending trillions on data centers. But they haven’t actually shipped models that could do this sort of intelligent compute.
It’s important to think seriously about this stuff but we literally have no idea. He refers to the 3-body problem as not being solvable by AI, and solving for AI economically is a similar thousand dimensional nonlinear system of differential equations in regimes we have never encountered before. We thought the Internet would empower democrats and it seems to have empowered autocrats.
I think some of Amodei’s conclusions are directionally wrong.
Society can’t process the current rate of change without massive disruption and unrest, never mind 100 years of innovation in 10 years.
If you automate 90% of people’s jobs, a huge number of people can’t do the 10% that remains that requires critical thinking etc., a lot of jobs are eliminated and wages go down. Hell, the job market is tough now and tons of people can’t handle the physical requirements as they get older and resort to pain pills, tons of people get aged out of tough jobs. I guess there are sort of Walmart greeter jobs.
The dependency rate goes up a lot if everyone lives to 100, people aren’t doing those jobs until they are 90. The human body hasn’t evolved for that, no one knows what happens if you fix all the stuff that would otherwise kill you, people will be like Tithonos who was granted immortality and just decayed infinitely without dying. Everyone will be taking care of their great-grandparents.
Follow the latest AI headlines via SkynetAndChill.com on Bluesky