Is Q* the A star pathfinding search combined with Q-learning, as claimed in a "tin hat time" tweet?
30
361
530
2025
27%
chance

https://twitter.com/natolambert/status/1727474191925182849

Doesn't have to be exact, just close enough to be informative

Get Ṁ200 play money
Sort by:
bought Ṁ35 of YES

I bought yes assuming that this will payout even if Q* turns out to include other insights

predicts NO

Yes, if Q* includes A*, Q-learning, and other stuff this pays out. "Doesn't have to be exact, just close enough to be informative"

predicts YES

@Alex my personal wild guess is that it's A* combined with a many billion parameter semantic search partner to guess at heuristic values for any search problem.

bought Ṁ100 of YES

Wouldn’t that basically be a muzero-esque algorithm? Mcts with pruning via your value function (the llm)

I'm against these kinds of markets. I don't think we should have markets that encourage the leaking of details about potentially dangerous capabilities advances.

predicts NO

@ChrisLeong I find it hard to believe that anyone will commit such a serious crime for mana when they could commit the exact same crime but sell the information to Google.