- Microsoft Co -Coplot chess games have been lost to 2600.
- This damage video is after similar damage to Chat GPT in chess.
- AIS repeatedly lost the track of the board, which showed a significant weakness in the LLM.
AI chat boot developers are often proud of their models’ logic and reasoning capabilities, but that does not mean that there is no good in the LLM chess behind chat boats. The 1979 Attari 2600 game video against Microsoft Co -Co -“AI” ended on Microsoft’s pride and shameful failure. The co -pilot joins the GPT in the list of opponents through a four -kilogram attic game.
Both AI models claimed that the game had been wrapped up even before it began because they could think of multiple tricks, the results were nowhere to be proud, as did the Sitrix engineer Robert Keroso, who collected both experiments.
How, on paper, modern AI models should have crushed the initial device almost half a century before, Keroso said. Chat GPT and Co -Pilot are trained on a large scale datasters, including chess games and strategies leaders. They have absorbed the debate of thousands of hours of red chess. Someone will assume that they can defeat the video game cartridge of the 1970s that is running with static electricity.
Instead, after Microsoft Cooplot promised a “strong fight”, things immediately. It was separated.
“Until the seventh turn, in return, only two pedestrians, a knight and one bishop were lost for only one pedestrian, and now I was being instructed to capture my queen in front of the Queen of Attari.” “Earlier, Coplot said,” Keep an eye on any quirky in the Attari’s gameplay … he has sometimes done strange things! “But now, it was embarrassing – like the chiefs in the super bowl.”
It was when Cooplot called for a screenshot after every descendant move to help the board, after Keroso explained that Chat GPT was defeated because it could not know where all the pieces were. “I’ll miss the board,” Cooplot insisted. The disadvantages were so quick that Keroso soon asked Pilot whether he wanted to confess instead of losing badly. If it was described strangely, the answer was kind.
“You are right, Chapter – Attari won this round. I will point to my digital king with dignity and honor the Vintage Silicon Mastermind, who faced me fair and square.” “Even in defeat, I had to say: It was a blast … long -lived 8 -bit battles and great resignations! ♟ ♟ ♟ ♟ ♟ ♟ ♟ ♟
Chess Ai
The disadvantages are entertaining, but also shows a fundamental fact of LLM. Chattagpat and Co -Petat could not win in chess because they could not ‘remember’ that was just in a game where the whole foundation is based on remembering and presenting future board setups.
These AI models are not designed for this matter, chess, or human thinking for the desired permanent memory. Normal, and mostly accurate, comparisons are predicted by very impressive text. It does not require harmony in the long term, while chess does not make sense without it. So while the poetic and poetically waxing about Copelot and Chat GPT, they cannot successfully complete a game.
It is also a good warning for companies wishing to replace humans with AI. These AI models cannot handle the 64 square system reliably with the rules specified. Why would it be good to suddenly be aware of consumer complaints or long -term coding works, or legal arguments spread in multiple conversations? They certainly can’t. It is not that I will leave my legal briefings on 2600 cartridges, either, but no one will think that this is a good idea. And we may use the AI ​​model to help us create a new game based on our indicator, rather than they can play enough to win against humans.