Gemini 4 Argon: Iteration, Games and Robot Control

Bijan Bowen37m 10s
0 comments · 0 votesOpen discussionClose discussion
Sign in to join the discussion

    Video summary

    Bijan Bowen tests Gemini 4 Argon through Antigravity on a browser operating-system mock-up, C++ skateboarding game and subway shooter. He finds useful interaction and gameplay logic, but crowded interfaces and modest visual detail. Video-based feedback adds mechanics and improves the games without consistently matching the polish of his earlier comparisons.

    Bijan Bowen reports that the model completes a robot-arm task in roughly 22 minutes, while still nudging the object during some movements. A Blender and Godot driving scene becomes much more detailed after several days of feedback, illustrating the difference between an early draft and a developed result.

    Bijan Bowen also demonstrates an older game running on Linux after a reported two-and-a-half-hour porting task. He treats this as an impressive practical result while acknowledging limited expertise for judging the generated technical explanation. The tests are his individual demonstrations, not controlled evidence of universal model superiority, exact binary equivalence or dependable autonomous deployment.

    Original YouTube thumbnailWatch on YouTube

    Share this page

    Bijan Bowen in a plain blue top against black, beside the blue and white headline “GEMINI NEEDS ITERATION”. Framed in blue with WWW.ARTIFICIAL-INTELLIGENCE.VIDEO, 13 October 2026 and duration 37m 10s.

    Bijan Bowen finds Gemini 4 Argon capable of demanding coding and visual tasks, but his hands-on tests show substantial iteration rather than consistently polished one-shot results.