Google’s most expensive AI model seems to have been a major milestone: hitting a 29 -year -old video game.
Last night, Google Sundar Pichai’s CEO of Posted triumphantly on x“What a finish! Gemini 2.5 Pro just finished Pokémon Blue!”
Be clear, the Gemini plays Pokemon Livestream Created by (in his own words) “a 30 -year -old mechanical software engineer not connected to Google” passing Joel z. But Google executives have checked the effort.
For example, Logan Kilpatrick, the Guide for Google Ai Studio, Posted last month This twin “made great progress in the completion of Pokémon” and had “won his 5th signal (the next best model has only 3 so far, although with a different braid agent)”, leading Pichai to joke“We work in API, Artificial Intelligence Pokémon :)”
Why Pokémon? In February, Manner underlined progress That Claude AI models did at “Pokémon Red”, writing that Claude’s “Extensive Thought and Training” gives him “a great boost” to “more unexpected” tasks, such as playing a classic game. (“Pokémon Red” and “Blue” are different versions of A game title Released for the first time in 1996 and was associated with the long franchise Pokémon). There is even A Claude Playing Twitch Pokemon channel That Joel Z is referred to as inspiration.
Despite his progress, Claude does not seem to hit “Pokémon Red” yet. Does this mean that Gemini is objectively better in the game? On his page, Joel Z urged viewers: “Please do not consider this point of reference for how well a llm can play Pokemon. You can’t really make direct comparisons – Gemini and Claude have different tools and get different information.”
Both AI models need help to play the game – there is where The aforementioned agent is exploiting Come, providing the models with snapshots of snapshots that overlap with additional information, allowing the model to decide how to respond (which may include the call of specialized factors) and then pressing the button corresponding to the AI instructions.
TechCrunch event
Berkeley, ca
|
June 5
Book now
Joel Z acknowledged that there were other “devings” to help the Gemini complete the game, but insisted that it was not cheating.
“My interventions improve the overall decision -making skills and logic,” he says. “I do not give specific tips – there are no walks or direct instructions for specific challenges such as MT. Moon. The only thing that is still coming close is to let Legtini know that he has to speak with a Grunt rocket twice to get the lift key, which was a mistake later in Pokemon.”
In addition, he said, “Gemini plays Pokémon is still actively developing and the frame continues to evolve.”
