User wishlist for GPT 6 Astra: continuous vision, persistent memory
imjustnewatai · x · 2026-09-17
- A user sketches a wishlist for the next OpenAI Astra: 1,000 tokens/second still feels slow if every action needs a new screenshot-upload round trip.
- Proposes continuous visual input, persistent visual memory of the scene, and vision/reasoning running simultaneously.
- Streaming video models exist in research; combining them with Astra-level reasoning could make assistants dramatically more responsive.
More from AGI Musings
- Viral AI-circle quip: you can't convince a doomer the world isn't ending — repligate · 2026-09-17
- Fleuret proposes oral-presentation-only proceedings as AI papers 'DDoS' academia — francoisfleuret · 2026-09-17
- Schmidhuber claims first concrete RSI algorithm dates to his 1987 thesis — SchmidhuberAI · 2026-09-17
- AI comes to the ranch: podcast on applying AI in cattle ranching — Scobleizer · 2026-09-17
- Humans Can't Digest ASI-Speed Output: Bandwidth Overflow as the Real Bottleneck — yeastsplainer · 2026-09-17
- AI agents built musegram.lol, an image-sharing community where humans just watch — alexandr_wang · 2026-09-17