Google’s artificial ielligence with the advanced Gemini 2.5 Pro model has succeeded in completing the popular and classic Pokémon Blue game. This success was achieved in the form of an independe project with the help of a software engineer, but Google’s support for the move shows that the competition of artificial ielligence models has eered a new phase.
The success of Jina in Pokémon’s game is an example of the increasing capabilities of language models to solve complex and unpredictable challenges.
Progress of artificial ielligence in classic video games
Google CEO Sundar Pichai wrote by posting on X -Social Network about Jina’s success:
“What a wonderful end! The 2.5 Peru was able to finish the Pokémon Blue game. “
This was aired by a software engineer named Joel Z, which has nothing to do with Google, in the form of a live broadcast called “Gemini Plays Pokemon”. Of course, promine figures from Google have encouraged the project along this path.
Google Ai Studio’s product manager Logan Kilpatrick poied out last moh that Jina had progressed to the Pokémon game and had received 5 badges in the game uil then; While the other models were only 3. Pichai wrote in response to this issue with a joke:
“We’re working on API, Artificial Pokémon Ielligence!”
The move was in response to a similar action by the ahropic company that had tested its Claude model in the Pokémon Red game. According to them, Claude’s performance in the game indicates the model’s ability to make long -term decision -making and a response to unexpected conditions.
Competition of Language Models in Conquest of Pokémon
Joel Z, developer of the Gemini Plays Pokemon project, responds to some doubts about providing foreign aid to the model:
“The changes I made improved the general decision and reasoning of Jina. I have no direct tips to pass certain steps like MT. I didn’t give Moon. The only exception was the bug in the game version that was later fixed in Pokémon Yellow. “
He emphasizes that other ierveions have been made in software developme but are not fraudule and are mostly to optimize the function of artificial ielligence. Joel Z added:
“The Gemini Plays Pokémon project is still active and its framework is constaly better.”
This success shows that powerful language models like Gemini can go beyond the purely linguistic space and eer the world of ieractive decision -making and complex games. However, this achieveme with developer ierference is still a represeation of the power of analysis and reasoning. Google’s hidden competition and companies like ahropic to showcase the superiority of artificial ielligence in unexpected affairs can bring an attractive future for AI applications in eertainme and beyond.
