Ask HN: Has anyone successfully played Tic-Tac-Toe with an LLM?

I tried to get both GPT-3.5 and GPT-4 to play Tic-Tac-Toe and both failed absolutely. Half the time they declared victory or defeat too early, made invalid moves, and let me cheat without questioning it. The other half of the time the moves they made seemed random and they always were easily defeatable. Has anyone else had similar experiences or have a prompt that has been more successful?