Bittensor miners score 42.3 on sports vision eval while GPT-6 Astra Ultra scores 0

markjeffrey · x · 2026-09-09

• Webuildscore's Bittensor subnet SN44 evaluated GPT-6 Astra Ultra on frame-level event detection over a 30s football clip, with ground truth revealed only after predictions were committed.
• The model scored F1 76.92% (precision 68.18%, recall 88.24%) on raw events, but 14.6 penalty points from three extra take-ons and two extra tackles plus misread timings pushed its decayed score to 0. The top Bittensor miner sits at 42.3.
• The poster notes this single 30s eval consumed 2% of his monthly pro credits, and publishes all work for public audit, claiming specialized miners now beat generalized LLMs at SOTA vision tasks.

Original post →

More from Fun

• Me every time AI comes up in casual conversation: the AI-fatigue meme — PurzBeats · 2026-09-09
• Indie maker Marc Lou auctions ad spots on 10 of his muscles, bids start at $1,000 — Polymarket · 2026-09-09
• GPT-5.6 Sol writing unit tests for a five-line change, meme edition — JeremyNguyenPhD · 2026-09-09
• First look at Andrew Garfield as Sam Altman in OpenAI drama film 'Artificial' — Krsn_Brkr · 2026-09-09
• ChatGPT as an interior designer: demo in action — zeeg · 2026-09-09
• Another 'Lee Sedol moment': model pulls an AlphaGo-style move — polynoamial · 2026-09-09

Fun channel →