Evaluating 35 open-weight models across three context lengths (32K, 128K, 200K), four temperatures, and three hardware platforms—consuming 172 billion tokens across more than 4,000 runs—we find that the answer is “substantially, and unavoidably.” Even under optimal conditions—best model, best temperature, temperature chosen specifically to minimize fabrication—the floor is non-zero and rises steeply with context length. At 32K, the best model (GLM 4.5) fabricates 1.19% of answers, top-tier models fabricate 5–7%, and the median model fabricates roughly 25%.
I refuse to call it AI
It’s a LM… Pure and simple. Anyway none of the LMs can come up with theory of relatively (if you gave them all of the known physics up to 1915).
Nor can they play paper scissors rock (they don’t realise it’s pointless).
As far as I can tell they’re wrong more times then they’re right and the only use I have for them is as a glorified search engine (and even then they’re still fricking wrong.
They’re only useful if you already know the answer because if you don’t know the answer you don’t know if they’ve given you the wrong answer.