AI such as Deep Blue was able to outperform humans at a specific tasks because humans wrote the algorithms.

This isn’t true of modern game-playing AI like Alpha-Go or recent Stockfish versions. These learn by playing against themselves over billions of games, and the strategies they develop aren’t guided by human input. These kinds of results aren’t achieved by LLM bots because there’s no equivalent to “winning the game” in a chatbot conversation that can easily be rewarded automatically.