In the preprint paper discussing ChatDev, the front page example is develop a gomoku game ( https://arxiv.org/abs/2307.07924 ). Given that, plus recent discussions of ChatGPT 3.5 Turbo Instruct's chess playing abilities (~1800 Elo is what I'm reading) I thought I'd give Turbo Instruct a spin on gomoku. Result? Awful. Won't block threes without prompting. Legal, but terrible moves. Told me it couldn't play Pro Pente as an AI language model. Sigh. Maybe someone wants to get ChatDev to make some Pente playing software? https://github.com/OpenBMB/ChatDev
Retired from TB Pente, but still playing live games & exploring variants like D, poof and boat
Posts:
2,232
Registered:
Mar 4, 2007
From:
San Francisco
Age:
45 Home page
Re: ChatGPT and gomoku
Posted:
Sep 24, 2023, 10:46 PM
If you get chat gpt successfully integrated into pente, it would be the final nail in the coffin of pente. Maybe dont pursue that watsu. At least mark prog cant identify post 3s. Let's not try to make things worse
Scire hostis animum - Intelligere ludum - Nosce te ipsum - Prima moventur conciliat - Nolite errare
Re: ChatGPT and gomoku
Posted:
Sep 27, 2023, 3:00 PM
Viktor Barykin already appears to have had a stronger program than Mark's in real time D Pente games back in 2007. https://pente.org/gameServer/forums/thread.jspa?forumID=1&threadID=3345&tstart=509 This is long before the AlphaZero program results. Certainly, a program would take little effort to create today which would beat Mark's program. Whether a chatbot software company could do it (create Pente or D Pente software which would beat Mark's program at 3 minutes per move) in 7 minutes of programming time for under a dollar, I doubt it at this point in time, anyway. Also, the D Pente contest between Mark's and Viktor's programs in 2007 in D Pente was first move to the center of the board.
Retired from TB Pente, but still playing live games & exploring variants like D, poof and boat