Among Us
A multiplayer online social deduction game in which players take the roles of crewmates or impostors, achieving victory by completing tasks or seeing through disguises. It is one of the representative titles of the genre.
Timeline (last 90 days)
The author believes the trained network was repeatedly calibrated over thousands of rounds in the same simulator, so its suspicion scores are highly aligned with who ultimately turns out to be the impostor, whereas the general knowledge brought by the language model offers no advantage in this environment, because the impostors' speech has no obvious anomaly and the only way to see through it is to check statements against who could see that room, but no results show the model can reliably perform that check.
Developer Seth Wheeler replaced the meeting voting system originally driven by a small recurrent network with the open-weight model gpt-oss:20b in an Among Us simulator, and the result was that across six configurations, the large model never made the crew perform better.
Among Us is a multiplayer social deduction game released in 2018, with 4 to 15 players per round, a minority as impostors and the rest as crewmates, divided into a free-movement phase and a meeting phase.
In the configuration with eight players and two impostors using second-league bots, the trained network's crew win rate was 52.0%, with an average of 0.99 impostors ejected per round, whereas gpt-oss in the turn-based Q&A format won only 2 of 16 rounds (12.5%), ejected 0.06 impostors per round, and wrongly ejected 0.94 innocent crewmates.
After adding the trained network's intuition to the eight-player configuration and using the turn-based Q&A format, gpt-oss as crew won 8 of 16 rounds, almost indistinguishable from the network baseline (p = 0.54).
In the strictly comparable experiment with fifteen players and three impostors in the sixth league, the trained network's crew win rate over 300 rounds was 93.3%, while gpt-oss under the dialogue+intuition format won 16 of 20 rounds (80%), with a one-sided Fisher exact test p = 0.052.
In the sixth-league fifteen-player rounds, meetings using gpt-oss took an average of 524 seconds of wall-clock time, with almost all the time spent on turn-based speaking across about 4 meetings, calling the model once per player per round; the trained network's voting requires no model calls.
Across the six configurations, the language model was clearly worse than the network only in the eight-player no-intuition configuration, and among the five with intuition it tied once and lagged four times, but each gap on its own was too small or had confounders and could not be conclusive; in none of the configurations did it perform better.
A social deduction game is a type of game that relies on information asymmetry and verbal confrontation among players to progress, with representative works including "Werewolf," "Among Us," "Blood on the Clocktower," and others.
"Duperie" is closer to a logic puzzle game with a social shell than to a party game like "Among Us."
5 more timeline events
All Facts (15)
The author believes the trained network was repeatedly calibrated over thousands of rounds in the same simulator, so its suspicion scores are highly aligned with who ultimately turns out to be the impostor, whereas the general knowledge brought by the language model offers no advantage in this environment, because the impostors' speech has no obvious anomaly and the only way to see through it is to check statements against who could see that room, but no results show the model can reliably perform that check.
50%UnverifiedDeveloper Seth Wheeler replaced the meeting voting system originally driven by a small recurrent network with the open-weight model gpt-oss:20b in an Among Us simulator, and the result was that across six configurations, the large model never made the crew perform better.
50%UnverifiedAmong Us is a multiplayer social deduction game released in 2018, with 4 to 15 players per round, a minority as impostors and the rest as crewmates, divided into a free-movement phase and a meeting phase.
50%UnverifiedIn the configuration with eight players and two impostors using second-league bots, the trained network's crew win rate was 52.0%, with an average of 0.99 impostors ejected per round, whereas gpt-oss in the turn-based Q&A format won only 2 of 16 rounds (12.5%), ejected 0.06 impostors per round, and wrongly ejected 0.94 innocent crewmates.
50%UnverifiedAfter adding the trained network's intuition to the eight-player configuration and using the turn-based Q&A format, gpt-oss as crew won 8 of 16 rounds, almost indistinguishable from the network baseline (p = 0.54).
50%UnverifiedIn the strictly comparable experiment with fifteen players and three impostors in the sixth league, the trained network's crew win rate over 300 rounds was 93.3%, while gpt-oss under the dialogue+intuition format won 16 of 20 rounds (80%), with a one-sided Fisher exact test p = 0.052.
50%UnverifiedIn the sixth-league fifteen-player rounds, meetings using gpt-oss took an average of 524 seconds of wall-clock time, with almost all the time spent on turn-based speaking across about 4 meetings, calling the model once per player per round; the trained network's voting requires no model calls.
50%UnverifiedAcross the six configurations, the language model was clearly worse than the network only in the eight-player no-intuition configuration, and among the five with intuition it tied once and lagged four times, but each gap on its own was too small or had confounders and could not be conclusive; in none of the configurations did it perform better.
50%UnverifiedA social deduction game is a type of game that relies on information asymmetry and verbal confrontation among players to progress, with representative works including "Werewolf," "Among Us," "Blood on the Clocktower," and others.
50%Unverified"Duperie" is closer to a logic puzzle game with a social shell than to a party game like "Among Us."
50%UnverifiedDisguisedToast accumulated a large fanbase through Among Us livestreams during the 2020 pandemic.
50%UnverifiedAmong Us is a social deduction game developed by Innersloth.
50%UnverifiedThe interior of the custom PC case is filled with Among Us character design elements, and the front bears the streamer's exclusive logo.
50%UnverifiedAmong Us was released in 2018 and went viral during the 2020 pandemic due to extensive streaming by a large number of content creators.
50%UnverifiedThe design philosophy of shorter matches is also reflected in similar social deduction games such as Among Us and Goose Goose Duck.
50%