Google’s most expensive AI model went beyond the main milestone to hit a 29 -year -old video game.
Last night, Google CEO Sundar Pichai won X.
Obviously, Gemini plays Pokemon Livestreamream.
For example, Logan Kilpatrick, a product lead for Google AI Studio, last month, Gemini is showing great progress in completing Pokémon, and has gained a fifth medium. 🙂
Why Pokemon? In February, the Claude AI model emphasized the progress created by “Pokémon Red”, and Claude’s “expanded thinking and agent training” recorded that it provides a “main boost” for “unexpected” work, such as playing a classic game. (“Pokémon Red” and “Blue” are other versions of the game boy title related to the Pokémon franchise, which was first released in 1996 and executed for a long time.) There is also a Claude Pokemon Twitch channel cited by Joel Z as inspiration.
Despite progress, Claude seems to have not yet beat the “Pokémon Red”. Does it mean that Gemini is objectively better in the game? Joel Z urged the viewer on his Twitch page as follows: “Do not take into account the benchmarks about how well LLM can play Pokemon. You can’t actually compare it directly. Gemini and Claude have different tools and receive different information.”
The two AI models also need to help play the game. The agent harness mentioned above appears, and a game screenshot containing additional information is provided to the model to determine the method of the model (how to call a special agent), and then press the button that corresponds to the instructions of the AI.
TECHCRUNCH event
Berkeley, California
|
June 5th
Please make a reservation now
Joel Z acknowledged that Gemini has another “Dev intervention” that helps to complete the game, but argued that it is not cheating.
“My intervention improves Gemini’s overall decision -making and reasoning ability,” he said. “I don’t give a specific hint. There’s no practice or direct guidance on certain challenges such as mountains after the door. The only thing that’s getting closer is to tell Gemini that I have to talk to the rocket twice. Later it was a fixed bug in Pokemon yellow.”
He also said, “Gemini is still actively developed by Pokémon and the framework is continuously developing.”
