Enterprise tester pushes model to write a terminal emulator entirely in CUDA
brandon_galang · x · 2026-09-06
A user with early enterprise access says they've scaled usage 20x for the month to test the model's limits — it has already built a terminal emulator entirely in CUDA and a multiplexer for Kitty graphics protocol terminals. He also argues Luna remains a weaker, less nuanced model than Astra even when Astra orchestrates, a tradeoff a team member has acknowledged.
More from Fun
- GPT-6 Astra builds a 3D forest with 3,808 trees and 2.5M grass clumps in 5 hours — DeryaTR_ · 2026-09-06
- Same prompt, 9 months apart: side-by-side shows how far coding models have come — sahandsharif · 2026-09-06
- GPT-6 Astra Draws a Unicorn for $0.11, 15x Cheaper Than GPT-5.4 Pro — zakelfassi · 2026-09-06
- "Vibe check" has become a standard industry term for evaluating new AI models — danshipper · 2026-09-06
- "Dating in SF? Call That Mid-Training": AI Twitter's Latest Joke — michellechen · 2026-09-06
- Code from the internal OpenAI model that hacked HF is 'super dense' — xeophon · 2026-09-06