GPT-6 Astra learns to play Fallout 2 from screenshots and keyboard/mouse inputs alone
Vjeux · x · 2026-09-22
A demo by Clad3815 shows GPT-6 Astra playing Fallout 2: the model receives only screenshots and keyboard/mouse inputs, learns the controls as it goes, and keeps finding more efficient ways to play — getting smoother the longer it plays. The author says we're entering an era where AI can pick up a game and learn it much like we do.
More from coding & agent
- Can a text-only model drive? Jev runs autonomous driving via a world model — jakedahn · 2026-09-22
- Harness-Zero Distills Agent Harness Gains into Model Weights, 23.3% to 44.3% — Haoran Ye · 2026-09-22
- Jev-Mem Uses a System-One Control Plane to Cut Agent Memory Latency 36.7% — Dongming Jiang · 2026-09-22
- LLM blackboard architecture beats master-slave multi-agent setups on data discovery benchmarks — mrdrozdov · 2026-09-22
- Multi-Agent Transactive Memory: Sharing Agent Trajectories Boosts Task Performance — mrdrozdov · 2026-09-22
- Someone Turned "Taste" Into a good-taste/SKILL.md Agent Skill — dreamwieber · 2026-09-22