Aaron Dharna
aadharna.bsky.social
PhD Student at UBC under Dr. Jeff Clune. Interested in RL, Generative Modeling, and Games (both for AI and for me)
👤 Joined on 23 December 2023
63
Followers
106
Following
29
Total Posts
📜 Kullanıcı Geçmişi
- 23 Ara 2023, 21:24
🎉 Bluesky'e katıldı - İlk kullanıcı adı:
@aadharna.bsky.social
🚫 Blokladığı Kullanıcılar
❤️ Beğendiği Gönderiler
We have tried! There’s a really nice paper that at least is predictive of policy gradient algos arxiv.org/abs/2312.08369
Show Post
that was six video releases in six weeks. I don't have any more in the can so it's back to the editing station for me. I hope I can get the next thing turned around reasonably quick. the answer to "can my channel survive after ending The Alt-Right Playbook and a year hiatus?" is a resounding maybe
Show Post
Lovely paper, long live to quality-diversity methods!
Show Post
Hmmm.. good point, but hard to say how constraining that is. We specifically ask it to produce policies in the form of code. So even if "must" be in its knowledge base, it might still be easier to discover the right exploits by expressing policies as code.
Show Post.. i.e. where a policy is a piece of code that pieces together known things and it's the composition of the components that make up the exploit, which could be hard to discover through RL alone.
Show Post