Speaking of human-level capabilities, it looks like I totally failed to correctly read the section of your comment that I quoted. Shame on me.
However, I'd expect that "Appearing to fail to reason well enough to know how to always fail to lose, and -if the opportunity presents itself- win at one of the simplest games there is." is absolutely not a desired outcome for OpenAI, or any other company that's burning billions of dollars producing LLMs.
If their robot was currently reliably capable of adequate performance at Tic Tac Toe, it absolutely would be exhibiting that behavior.
However, I'd expect that "Appearing to fail to reason well enough to know how to always fail to lose, and -if the opportunity presents itself- win at one of the simplest games there is." is absolutely not a desired outcome for OpenAI, or any other company that's burning billions of dollars producing LLMs.
If their robot was currently reliably capable of adequate performance at Tic Tac Toe, it absolutely would be exhibiting that behavior.