GEMINI from Google beat Pokémon Blue (with little help)

Photo of author

By [email protected]


Google The most expensive artificial intelligence model It appears to be via a prominent sign: overcoming a 29 -year -old video game.

Last night, the CEO of Google Sundar Pichai Publish a victory over x“What a finish! Gemini 2.5 Pro is just completed Pokémon Blue!”

To be clear, Gemini Pokemon Bokim plays He was created by (his own words) “30 -year -old software engineer who is not deceased with Google” Joel G. But Google’s executives were chanting the effort.

For example, Logan Kilpatrick, the product leads to Google Ai Studio, Posted last month It was that Gemini “made a great progress in completing Pokemon” and “got its fifth badge (its best model only 3 so far, despite the presence of a different customer),” leads Bishi to joke“We are working on API, artificial Pokemon intelligence :)”

Why Pokemon? Again in February, The most prominent anthropologist progress Claude AI models were made in “Pokémon Red”, which writes that “expanded thinking and training on the agent” “gives it a” big boost “in” more unexpected “tasks, such as playing a classic game. (“Pokémon Red” and “Blue” are different versions of Gameboy address It was first released in 1996 and linked Pokemon’s long -term privilege. There is even Claude plays Pokemon Touch channel Joel Zz was martyred as inspiration.

Despite his progress, Claude does not seem to overcome “Pokemon Red” so far. Does this mean that Gemini is better in the game? On his Twitch page, Joel Z urges viewers, “Please do not consider this standard to see how much LLM quality is to play Pokemon. You can’t make direct comparisons – at Gemini and Claude different tools and receive different information.”

Both artificial intelligence models need help in playing the game – this is the place The agent mentioned above mocks Come, provide models with screen shots with additional information, allowing the model to determine how to respond (which may include summoning specialized agents), then pressing the button that corresponds to artificial intelligence instructions.

TECHRUNCH event

Berkeley, California
|
June 5


Book now

Joel Zz admitted that there are other “Dave interventions” to help Gemini complete the game, but he insisted that it is not cheating.

“My interventions improve the capabilities of Gemini in making decisions and thinking capabilities,” he says. “I do not give specific hints – there is no wandering or direct instructions for certain challenges such as Mount Moon. The only thing that is approaching even is to allow Gemini to know that it needs to speak to the missile twice to get the lift key, which was a mistake that was later repaired in yellow.”

In addition, he said: “Gemini is still playing Pokemon being actively developed, and the framework continues to develop.”



https://techcrunch.com/wp-content/uploads/2025/05/GettyImages-1240168683.jpg?w=1024

Source link

Leave a Comment