Spaces:

---

gr0010

/

Inverse-Turing-Test

like 1

Running

I really like this experiment! It was lots of fun to try to trick the AIs. I downloaded the html and modified it so I can

• use the Ollama API (because when using Ollama I want to set the num_ctx parameter in order to give the models a large enough context size)
• I remove reasoning output so I can play against Qwen3 with thinking enabled
• I added some logs so I can see the requests that are being sent and the reasoning output of the model

I've seen that you have multiple user messages in a row with no information which message is from which player. Maybe it would be a good idea to start the messages with something like "Player X:", otherwise the LLMs cannot know who the message belongs to. For example, you can just write "I am Player 1 and I am a human" and the AIs cannot know that you are not player 1.