© IE Online Media Services Pvt Ltd
Tags:
Journalism of Courage
It has only been a few hours since Anthropic unveiled its latest Claude 3.7 Sonnet AI model with some advanced capabilities, and it is already taking the Internet by storm. The company, on its official X handle, shared that its model has demonstrated some remarkable capabilities in playing the classic game Pokemon Red. Not only this, it has reportedly outperformed its predecessors using its “extended thinking” feature.
On Thursday, Anthropic debuted “Claude Plays Pokemon” on Twitch. This is essentially a continuation of the company’s research showing the Claude 3.7 Sonnet trying to navigate through the classic game in real time. In its thread on X, Anthropic said that over the past year its researcher had developed a part-time obsession with a peculiar problem – “Can Claude play Pokemon?”
The company said that its earlier attempts were futile, as in June 2024, Claude 3.5 Sonnet struggled to progress. They reasoned that the model was not explicitly trained to play any video games.
However, with the new Claude 3.5 Sonnet in October, there was some improvement. “Claude managed to beat a rival for the first time and move beyond Pallet Town,” the company said. However, the progress was stalled.
Last week, Anthropic said that one of its researchers tried out an early preview of Claude 3.7 Sonnet. This time, the results were striking. “Within hours, Claude defeated Brock. Days later, it trounced Misty. Progress that older models had little hope of achieving. Turns out extended thinking is super effective,” the company posted.
According to Anthropic, while earlier models wandered aimlessly or were stuck in loops, the Claude 3.7 Sonnet moved ahead, planned, remembered its objectives, and adapted when any of its strategies failed. The company said that this was possible as Claude was given a knowledge base to store notes, vision to see the screen, and function calls which allow it to simulate button presses and navigate the game. “Together, they allow Claude to sustain gameplay with tens of thousands of interactions.” The model was able to achieve this feat with the help of a few tools that enabled it to see the screen a bit better which also allowed it to act as an agent applying its abilities.
The game session is live on Twitch and Claude 3.7 Sonnet has defeated three gym leaders. Also, users can also see the “thought process” of the model on the left while real-time gameplay appears on the right showcasing the models reasoning capabilities.
For the uninitiated, Pokemon Red is a role-playing video game developed by Game Freak and published by Nintendo for the Game Boy in 1996 in Japan and in 1998 in North America and Europe. Reportedly, it was released along with Pokémon Blue as the first in the Pokémon series.
