Prometheus awarded $25,000 in prizes, generously funded by the FTX Future Fund.
Successful Proposals: $5,000-$50,000
We may undercut or exceed our budget based on the quality of submissions. To ensure a fair commitment, Prometheus will commit to awarding at least $25,000 in total to the most novel proposals we receive even if very few proposals meet the criteria below. We may also increase the prize pool above $100,000 if we find proposals far above our expected quality.
Each successful proposal will receive at least $5,000, where “successful proposal” is defined using the criteria below. Proposals which are especially promising or present novel ideas or approaches can win up to $50,000.
We want to encourage collaboration on this project, as multiple minds often work far better than only one. For proposals submitted by groups of 2-3 people, EACH member will receive ⅔ of the prize that the team would have gotten—for more total prize money for the team.
Award Criteria
There are two ways to win a prize in Prometheus ELK. Your proposal should clearly acknowledge by which method you are attempting to win a prize. We expect that proposals which are entirely new are substantially more difficult to create than proposals that build off existing ones, but are also likely to win larger awards.
The first way to win a prize is to develop a specific training strategy for ELK that handles ALL of the counterexamples summarized here. In other words, a winning strategy would require a new counterexample or something new about the test case to cause their strategy to break down. You do NOT need to fully solve the problem, or make any sort of significant breakthrough.
The other, likely more approachable, way to win a prize is to take a past proposal and extend it to cover the new counterexample. During ARC’s ELK contest, a number of people submitted successful proposals not covered by existing counterexamples, and ARC showed how to defeat these proposals in their prize results announcement. If you successfully take a proposal that has won a prize and modify it in order to defeat all original counterexamples, as well as the new counterexample it was broken by, you’ll win a prize.
Submit your proposal.
You can submit your finished proposals and/or questions to elk@prometheus.science. Smaller questions can also be asked in our discord server at discord.gg/prometheus – if you are new to the server, you may need to wait 24 hours after registering to be verified.
A basic outline for your proposal can be copied from our template, and you can modify it in any way you wish as long as you include all details. If you submit a google document, make sure that we have access, or your submission will not be assessed. Submissions will be assessed by Devansh Pandey and Mark Xu, along with potentially other researchers already familiar with ELK. Best of luck!