Popverse's top stories

• Discover (or Rediscover!) the origins of the Critical Role's Mighty Nein before Amazon's Prime Video TV series in a new boxed set coming this fall
• MEMBERS ONLY: Popverse Picks: The best Stephen King movie & TV (and more) adaptations, from one King fan to another
• Netflix's next big hit anime The Summer Hikaru Died goes hard into cosmic horror

You can say what you want about AI models like ChatGPT – and we do because they’re pretty terrible – but at least they should be good at finding the most logical route through a game like Pokémon. Not so, apparently, as OpenAI, the company behind ChatGPT, has put its O3 model to the test playing Pokémon Red. Almost 150 hours in and it has only just gotten the first three badges, putting it well behind the curve for similar playthroughs on Twitch.

Considering it can’t do anything as well as a human, we know that we shouldn’t expect an AI to be able to play a video game as well as a human. Still, there is something unintentionally hilarious about watching an AI struggle to play a game that the entire Twitch community, doing seemingly random inputs, took just short of 400 hours to beat back in 2014. At the time of writing, the AI was still struggling to get through Rock Tunnel to Vermillion City, meaning that it still has most of the badges, plus the Elite Four to get through before it can claim victory. Even with a massively over-leveled Nidoking at the head of their party, they’re struggling to make any progress in the Rock Tunnel.

As a turn-based game with minimal puzzles, an AI should be well-suited to getting through Pokémon Red. However, even O3, which is designed to be a problem-solving algorithm, is struggling with the game at the moment. Unless it picks up the pace pretty significantly in the second half of Pokémon Red, we don’t expect it to beat humanity in the gaming sphere just yet.

You don't need to beat the game to prepare for the next one—here are all the major new and upcoming games coming our way.