3,593 karma · joined July 30, 2007
I'm applying multi-agent reinforcement learning to AI Safety. I want to ensure that Artificial Intelligence will benefit human society rather than hurt it. I'm addressing both short-term consequences of existing AI tools such as LLMs and long-term existential risk imposed by the potential development of Artificial General Intelligence.
I propose to solve the problem of AI Safety by improving the social behavior of AI agents. Most AI research is focused on getting AI agents to solve bigger and more complex problems in the most efficient way possible; my research is focused on achieving cooperation, reciprocity and teamwork between AI agents.
My strategy is to distill lessons from the study of animal and human societies and apply them to AI agents.
Read more about my research here: https://r.rachum.com
Maybe 50-year-old COBOL programs are the original neural networks.
$ posh /home/ramrachum/Dropbox/notes.txt
$DX/notes.txt
Of course, it only becomes useful when you define a bunch of environment variables for the paths that you use often.I use this a lot in all of my scripts. Basically whenever any of my script prints a path, it passes it through `posh`.
Apparently no one else was interested in it. I still use it daily.
> omg, this sounds like the gigantic, ossified and crushing bureaucracy of a third world country.
No, it sounds like how most successful organizations work.
Software engineer with 12 years experience. PSF Fellow and Ex-Googler.
Anything Python-related. Bonus points if I get to work with JAX.
I'm currently doing multi-agent reinforcement learning research using RLlib, which is part of Ray. I tried to install a PyPy environment for it. It failed because Ray doesn't provide a wheel for it:
Could not find a version that satisfies the requirement ray (from versions: none)
My hunch is that even Ray did provide that, there would have been some other roadblock that would have prevented me from using PyPy.