Jason Gorman @jasongorman.bsky.social · 05/10/2026OpenAI accuses a rival Chinese lab of "stealing" "hidden reasoning" by launching thousands of queries against their models. That's quite revealing about what LLM "reasoning" actually is. 592
Jason Gorman @jasongorman.bsky.social · 05/10/2026Imagine they built chess models, and accused a rival of stealing their model's "chess reasoning" by studying its output. That would suggest the model was trained on a large number of chess games and is just predicting which move's most likely to come next. Which, BTW, is exactly how LLMs play chess 110
Jason Gorman @jasongorman.bsky.social · 05/10/2026If you play an LLM at chess, you'll see by the 15-20th move, as the sequence goes out of the model's training data distribution, that they're not reasoning about the game at all. 110
