Current community forecast
Yes 56.2%
Forecasters
12
Question type
binary
Methodology
Play-money forecasting platform
Source type
Forecast
Market data
Updated 36 minutes ago
Sep 20, 23, 8:10 PM UTCDec 31, 27, 10:59 PM UTC
Trends
Rules
General purpose (not trained for a specific task) language models demonstrated chess playing ability.
- They are also capable of deception and lie detection.
- Will language models or visual-language models* beat the average score of human players during a series of 40 games on webDiplomacy.net by 2028? (question modeled after Meta's Cicero result). [EDIT: Please notice that while "CICERO achieved more than 2x the average score of its opponents" this question requires only achieving the above-average score]
- *models or agents trained on different modalities (so e. g. models capable of controlling robotic arm like PaLM-E) would also qualify as long as they weren't trained specifically to play Diplomacy
