Google The most expensive AI model It seems that an important milestone has crossed: the 29 -year -old ancient video game defeat.
Last night, Google CEO Sunder Pachi Winner Posted on X“What finish! Gemini 2.5 Pro just completed the Pokémon Blue!”
To be clear, Gemini Pokémon Plays Live Stream (In your own words) was created by “a 30 -year -old software engineer” non -associated with Google. Joel Z. But Google executives are cheering this effort.
For example, Product Lead for Logan Call Patrick, Google AI Studio, Posted last month She “was making great progress in completing Pokémon” and she “received her 5th seed (3 in the next best model so far 3, though with the power of a different agent),” Joke“We’re working on API, artificial Pokémon Intelligence :)”
Why Pokémon? Back in February, Anthropic highlighted the progress Its cloud AI model was making in “Pokémon Red”, writing that Claude’s “expansionist thinking and agent training” gives it “an important promotion” on “more unexpected” tasks like playing classical games. (“Pokémon Red” and “blue” are different versions A Game Boy title The first release was released in 1996 and linked to the long -run Pokémon franchise). Here’s A Claud Pokémon Plays Tweet Channel That Joel Z cited as an inspiration.
Despite its progress, Claude has not yet defeated “Pokémon Red”. Does this mean that Gemini is objectively better in the game? On your Twitter page, Joel Z called the audience, “Please do not consider how well LLM Pokémon can play. You can’t really compare directly – Gemini and Claude have different tools and get different information.”
And both AI models need help to play games – the same place Uses the aforementioned agent In this, presenting models with the game screenshots with additional information, allows the model to decide (which includes calling special agents), and then press the button that is compatible with the AI’s instruction.
Taxkarnch event
Berkeley, ca
|
June 5 June
Joel Zeed acknowledged that there were “giant interference” to help Gemini complete the game, but he insisted that it was not a fraud.
“My intervention improves Gemini’s overall decision -making and reasoning capabilities.” “I don’t give specific indicators – there are no walkthroughs or direct instructions for special challenges like Mount Moon. The only thing that comes near is to tell Gemini that he needs to talk to rocket gratte twice to get the lift key, which was later set in Pokemon yellow.”
In addition, he said, “Gemini plays the role of Pokémon, it is still actively developing, and this framework is being developed.”