Data labeling isn't dead: cherry-picked demos ≠ long-tail superiority
chris_j_paxton · x · 2026-09-09
Responding to the viral claim that "data labeling is dead" (backed by a GPT-6 Astra demo distinguishing Celtics from Knicks players home or away), Chris Paxton cautions that cherry-picking a good example is far easier than demonstrating superiority on the long tail. He argues data labeling will last "as long as humanity does" — a pushback against over-reading flashy single demos.
More from Models
- Community cuts MiniMax H3 from 28 to 6 steps, topping the H3 Acceleration Arena — AdinaYakup · 2026-09-09
- Bayesian model selection explains why proxy-scale winners can lose at target scale — BlackHC · 2026-09-09
- Even if open models solve Navier-Stokes someday, nobody will credit them, due to assumed training data contamination — teortaxesTex · 2026-09-09
- Zvi teases Astra system card write-up, worries 'the world is inside my OODA loop' — TheZvi · 2026-09-09
- GPT-6 Astra tops RSI-Exam at 0.5126, 18.4% above GPT-5.6 Sol — HuaxiuYaoML · 2026-09-09
- Qwen beats ChatGPT at his tasks, now he hunts a UI to run it on whole code projects — yeah280 · 2026-09-09