In early 2027, I will play Minecraft for an hour with a friend over Discord, on a new server. I will attempt to do the same with an AI using the best publicly available scaffolding, such that I can speak in real time with an AI who is simultaneously controlling a Minecraft Agent. In particular, my conversation should have a causal effect on how the Agent is controlled.
This market resolves YES if I find my hourlong session with the AI at least 50% as enjoyable, entertaining, and interesting as the one with my friend, judged subjectively. I will make my determination based both on in-game events as well as the quality of conversation.
1,000People are also trading
I've tried this with Gemini (just the chatting while playing Minecraft part) and have found it technically a lot better than last year in hearing and stopping/starting appropriately, but it felt like speaking to a really knowledgeable genius who secretly despises me but can't express it, so instead just does whatever it can to finish conversations as quickly and perfunctory as possible.
I'd love to see a market covering just the conversation quality here. I don't think even that is >50% probable.
@DylanRichardson well that specifically seems at least partially solvable, considering AIs often already tend to be unnecessarily verbose and also more recent have been trained to keep the conversation going by asking questions etc
@AdamK so currently, it can't even do it at all, let alone do it well, right? How long before you'll try?