Question regarding atari games

[hlsafin]

In your opinion, can this approach be applied to Montezuma's revenge enviroment?

[conglu1997]

Hi, that would certainly be an interesting extension! One path would be to use GPT-4o’s vision capabilities and operate IGE on visual observations. Another route could be to summarize visual observations into text with GPT-4. Both options seem feasible to me :)