Google’s most expensive model seems to have overcome a milestone: Beating a 29-year-old video game.
Last night, Google Director General Sundar Pichai triumphed on X, “Conclusion Affairs! Gemini 2.5 Pro just finished Pokémon Blue!”
To be clear, Gemini Lion Pokémon Livestream was created by (in his words) “a 30 -year -old Google incompetent software engineer” who goes from Joel Z. but Google executives have cheered efforts.
For example, Logan Kilpatrick, the product lead for Google Ai studio, posted last month that Gemini was “making great progress in the end of Pokémon” and had “earned his 5th symbol (the other best model has only 3 so far, though with a different agent armor),” Pichai leader to joke, ” Intelligence: 🙂 ”, though with a different agent)”
Why pokemon? Back in February, anthropic emphasized the progress that his Claude models he was making in “Pokémon Red”, writing that Claude’s “Thinking and Training” gives her “a big stimulus” in “more unexpected tasks” as playing a classic game. (“Pokémon RED” and “Blue” are different versions of a Gameboy title first released in 1996 and associated with long pokémon exclusivity). There is also a Claude playing Pokémon Twitch Channel that Joel Z quoted as an inspiration.
Despite his progress, Claude does not seem to have beaten “Red Pokémon” yet. Does this mean that twins are objectively better in the game? On his Twitch page, Joel Z urged viewers, “Please do not consider this a landmark how well a llo can play Pokemon. You really can’t make direct comparisons – Gemini and Claude have different tools and get different information.”
And the two models of he need help to play the game – this is where the aforementioned agent’s harness enters, providing patterns with games coated with additional information, allowing the model to decide how to answer (which can include the specialized agents’ call), and then press the button corresponding to it.
Techcrunch event
Berkeley, ca
|
June 5
Reserve now
Joel Z admitted that there were “other interventions” to help the Gemini finish the game, but insisted he is not cheating.
“My interventions improve the overall decision -making and reasoning of Gemini,” he says. “I do not give specific suggestions – no ways or direct instructions for specific challenges like MT. Moon. The only thing that approaches even nearby is letting the Gemini know that he should speak with a strict rocket twice to get the elevator key, which was a mistake later fixed in the yellow pokemon.”
Plus, he said, “Gemini Luan Pokemon is still actively developing, and the frame continues to evolve.”
