1,443 karma · joined April 13, 2014
scientist.wang
So I'd wager the reason there was backlash against the bill was that it did not close the exemption, not (as mentioned in the article I posted originally) that it introduced the exemption.
topics: maybe some more awards for deep learning (Goodfellow?, DeepMind team for AlphaGo/Zero: Hassabis, David Silver, etc? If the current trend of transformers being hot continues, maybe some authors of "Attention is all you need"? Whoever makes the breakthrough that enables the singularity in the next decade?), a quantum computing cohort could win one year (Shor? Aaronson? idk anything about this field... Umesh Vazirani??).
edit: In an alternate universe where blockchain was used only for actually practical purposes instead of the current universe in which its invocation evokes images of cryptocurrency tiktokers, get rich quick schemers, and buzzword-spouters, in an alternate universe where the set of people interested in blockchain was very nearly the same set of people interested in, say, RSA encryption or zero-knowledge proofs, and if pseudonymous people could win (perhaps posthumously?? ok, this is a lot of conditionals...), satoshi would have to win one year, right? Maybe along with their precursors? Dunno if that usually happens with Turing awards.
I wonder if perhaps this opinion of Feynman as a showman has gotten more popular over the decades as Feynman and the physicists who knew him passed away, and more and more of us are exposed to him only through his lectures and books. (If you'll excuse an aside: a bit like Bill Simmon's thesis on the basketball greats Bill Russell vs. Chamberlain -- those of us looking back on history, with only artifacts, may draw conclusions that would be ridiculous at the time.) I don't know if this is the case, but it's a fun speculation.
https://www.atomicarchive.com/history/manhattan-project/p1s2...
I guess you could phrase it that way, but that's essentially the problem statement for developing a strategy for an imperfect-information game. So I would say it is a flaw in the principle if their final output is exploitable.
The bigger issue in my eyes was that while OpenAI 5 defeated the world champion team OG, when they let anyone in the world fight it, some ingenious players figured out a pretty robust method to consistently exploit and defeat the bot. As I haven't heard any buzz about OpenAI 5 since then, I think it was more or less unsuccessful unless they can show that their training method produces unexploitable bots (instead of bots that are really good against certain strategies)
I don't think I've been exposed to it, but I would love to be. Where did you learn about TCS history?
Previous discussion on the good article: https://news.ycombinator.com/item?id=16494280
Honestly, the currently linked article seems to me like low-effort blogspam.
6
2 1.0
3 1.0
4 1.0
3 0.5
5 0.678
9 0.87
First line means there will be 6 test cases. Each of the following 6 lines contains one test case (each test case represents one ping-pong tournament, and each test case is independent of the others).So the first line means there's a tournament with N=2 players, and P=1.0, which means that in any matchup, the higher-ranked player has a 100% chance of defeating the lower-ranked player.
The fourth line means there's a tournament with N=5 players, and P=0.678, which means in any matchup, the higher-ranked player has a 67.8% chance of defeating the lower-ranked player.
This reminds me of a recent problem in Facebook's annual coding contest (hacker cup), where the insight is similarly to notice that you only need to characterize states by #s instead of by the specific set of players: https://www.facebook.com/codingcompetitions/hacker-cup/2020/...