Very interesting! Did you really pre-compute the entire conversation tree? I assume it's all canned. In any case I found guessing randomly and carefully gave me similar result because I was so far off the desired thread by conversation end either way. Maybe would be better with shorter conversations.
Viewing post in Next Token Please jam comments
Thank you for playing and for the feedback! Not pre-computed, prompts and answers are generated as you play by a language model running locally in the backend, and the game turns them into playable rounds.
Fair feedback on conversation length too. Once you miss a few early picks the answer drifts so far that the rest feels random either way. Shorter answers could definitely help - it's something I want to experiment with to improve the gameplay.