Where the previous models toured without a goal or stumbled in the episodes, Claude 3.7 Sonnet plans forward, remember their goals, and adaptation when the initial strategies fail.
Critical skills of the gym leaders. We assume, to solve the problems of the real world as well. pic.twitter.com/scvisp14xg
Anthropicai February 25, 2025
“When this good strategy derives, I do not think it necessarily has self -awareness to know that strategy. [it] I reached better than the other. “This is not a trivial problem to solve it.
However, Hershey said he sees a “low fruit” to improve Pokémon toys from Claude by improving the style understanding of Game Boy shots. “I think there is an opportunity to overcome the game if it has an ideal feeling of what is on the screen,” Hershey said, saying that such a model is likely to be “slightly less than a person.”
Hershey said: The expansion of the context window for the Fuod’s future models may also allow these models “thinking about long time frames and dealing with things more coherent over a long period of time.” Future models will improve by getting “a little better in remembering, and tracking a coherent group than he needs to try to make progress,” he added.
Whatever you are thinking about imminent improvements in artificial intelligence models, Claude’s current performance in Pokemon does not make it look like he is ready to enter into the explosion of artificial intelligence at the human level. Hershey allows watching Claude 3.7 Sonnet stuck on MT. Moon for 80 hours or so can make him “look like a model that he does not know what to do.”
But Hershey still admires the way the Claude’s new model will appear to think sometimes some ray of consciousness and “somewhat he says he does not know what he is doing and knows that he should do something different. The difference between” cannot do it at all “, and” he can do it “. Really good. “