Azath Vitr (D, on 20 July 2026 - 11:36 PM, said:
Illuyankas, on 20 July 2026 - 11:02 PM, said:
As a specific counterexample it's very easy to formally verify. And yes, actual mathematicians have confirmed it.
As for "lying slop machine": Flagship models exhibit PhD-level intelligence across multiple fields. For mathematical proofs their logic can be automatically formally verified in Lean. But for a counterexample like this a few simple calculations are sufficient.
I've been using DeepSeek extensively in my study of graduate level pure mathematics. I always check its work rigorously, and it rarely makes mistakes. Research mathematicians have confirmed the extraordinary usefulness--and even the extraordinary creativity---of LLMs that have been enhanced with reasoning models.
Yeah ... don't use DeepSeek.
It may be a decent model by a lot of accounts, but it's even worse than the others for stealing your data and tracking you etc because this one is straight outta Beijing.

Help













