AI Companies Are Buying Rare Books, Scanning Them for Training Data, Then Shredding Them
Multiple AI companies are buying rare books, scanning them for training data, then shredding the originals — including irreplaceable texts.
JuliaHub Benchmarks GPT-5.6 Against Claude Fable 5 on Physical AI Tasks
By Prompt AI News1 min read
#openai#anthropic#benchmark#robotics
Per a post surfaced on Hacker News, JuliaHub has published a benchmark evaluation comparing OpenAI's GPT-5.6 and Anthropic's Claude Fable 5 on physical AI tasks — a category covering reasoning about real-world environments, robotics, and simulation.
The benchmark's scope goes beyond the text, code, and reasoning tasks that have defined most frontier model comparisons to date. Physical AI evaluation asks models to reason about how objects move, how robots should act, and how simulation environments behave — a meaningfully different test than standard language benchmarks.
The specific results from JuliaHub's evaluation were not available in the materials at the time of this report. The full evaluation lives on JuliaHub's site; the significance of this benchmark will become clearer once those results are reviewed directly.
JuliaHub's choice to test the two latest frontier models on physical tasks signals that this category is becoming a serious axis of model capability measurement. Robotics and embodied AI developers will be watching the results closely.
Leave a Comment
All comments are reviewed before appearing. Keep it respectful.
More in Commentary
View all →Instagram's Head Says Engineering Hiring Has Changed — Coding Fluency Is No Longer the Bar
Adam Mosseri says Instagram no longer requires engineers to code 40–60% of the time and has scrapped the traditional technical interview loop.
Sam Altman Says AI Won't Shorten the Workweek — Because People Enjoy Being Busy
Sam Altman says AI won't shorten the workweek because humans enjoy staying busy and will just create more work to fill the time.