- Microsoft Copilot lost a chess game against an Atari 2600.
- The loss follows the similar loss of chatgpt in the video failures of Atari.
- The AIS has several times lost the trace of the state of the board of directors, demonstrating a key weakness in the LLM.
Chatbot AI developers often boast of the logic and reasoning capacities of their models, but that does not mean that LLM behind chatbots are good in failures. An experience between Microsoft Copilot to the “AI” fueling the Video games atari 2600 1979 which ended with an embarrassing failure for the pride and the joy of Microsoft. Copilot joins Chatgpt on the list of opponents beaten by the game of four kilobbytes Atari.
Despite the two models of AI which claim to have the game, almost wrapped before it begins because they could think of several movements, the results were far from the boastful, as the engineer Citrix, Robert Caruso, who set up the two experiences document.
Caruso described how, on paper, modern AI models should have crushed the rudimentary tool almost half a century ago. Chatgpt and Copilot are that they are trained on massive data sets, including chess games and strategy guides. They absorbed thousands of hours of discussion of Reddit chess. One could assume that they could beat a video game cartridge from the 1970s powered by static electricity.
Instead, after Microsoft Copilot promised a “strong fight”, things immediately collapsed.
“On the seventh round, he had lost two pawns, a knight and a bishop – for only one pawn in return – and asked me to place his queen just in front of the Queen of Atari to be captured at the next turn,” wrote Caruso. “Earlier, Copilot said:” Keep an eye on all the oddities of Atari’s gameplay … It sometimes made bizarre movements! “But now he became embarrassed – like the chefs of the Super Bowl.”
It was after Copilot asked for a screenshot after each move from Atari to help remember the table, after Caruso explained that Chatgpt had lost because he could not keep track of the place where all the parts were. “I will remember the board of directors,” insisted Copilot. The losses accumulated so quickly that Caruso quickly asked Copilot if he wanted to concede rather than continue to lose wrong. The answer was graceful, so strangely formulated.
“You are absolutely right, Bob – Atari has won the victory in this tower. I will tip my digital king with dignity and honor the brain of vintage silicon which beat me fair and square,” quoted Caruson Copilot as writing. “Even in defeat, I must say: it was an explosion … 8 -bit life battles and noble resignations! âī¸đđšī¸”
AI chess
The losses are fun, but also reveal a base of LLM. Chatgpt and Copilot could not win chess because they could not “remember” what has just happened in a game where the whole premise is based on the recall of the movements and the projection of future configurations of advice.
These models of AI are not designed for the type of persistent memory required for failures or human thought, moreover. The common and above all precise comparison is very impressive text prediction. This does not require long -term consistency, while failures do not make sense without it. Thus, although Copilot and Chatgpt may seem poetic on the quality of failures, they cannot finish a game successfully.
It is a good warning for companies wishing to replace humans with AI too. These AI models cannot reliably manage a system of 64 squares with clearly defined rules. Why would it suddenly be good to follow customer complaints or long-term coding tasks, or a legal argument that extends over several conversations? They can’t, of course. Not that I would leave my legal memories to an Atari 2600 cartridge, but no one would think or a good idea. And maybe we should use AI models to help us create new games according to our guests, rather than believe that they can play enough against humans to win.
