Am a college student at MIT in Math and CS. Planning for what I'll do next summer, give suggestions! (Might increase bounty). Website: https://uzpg.me
Background info:
into ML/AI stuff, did research on theory of deep learning and scaling
for a long time I've been into building software to help people learn and come up with ideas, and last summer I worked at dust.tt on LLM interfaces. I think I want to pursue more research style thinking next summer, but I'm open to stuff.
enjoy mathy stuff a lot right now
Other interests: functional programming, cybersecurity, lit/philosophy, physics
see website ^
Considering:
working at a research lab at MIT or some other university (this shouldn't be too hard?)
applying to some AI labs for research internships
playing around with product ideas and doing a startup surrounding some cyborg/AI enhanced learning stuff with a friend
more mathy, physics or theory of CS research if I find good mentorship
vibes, learn shit and go around a bit
Appreciate the help! If people have good takes Ill up the bounty, last time I just started too high.
Do an AI interpretability experiment to measure levels of feature linearity and test how they change based on model size and architecture: https://www.lesswrong.com/posts/ziNCZEm7FE9LHxLai/?commentId=rAJ5goAkdwknSekDv
I had the idea for this experiment last month, and a friend who's a CS grad student came up with the method that's further down the comment chain, but I don't have the time to do it myself. If the experiment is successful, it could provide very useful information about the scalability of approximate-linearity-dependent alignment techniques.
You could look at some of the government AI jobs, some of them open in a couple days: https://ai.gov/apply/
If you like climbing, you could try bouldering.
For a combination of music and AI, you could listen to Spotify's DJ and learn more about how it works or how you could make something like that.