Dumb question but with something like 15 million lines of Lean code, how can we tell when the Lean build “passes” that it is actually formally verifying the right thing and there wasn’t just a “1+1=2” making it pass? Like you hear people say “if your Rust program compiles then it’s correct” but that never meant your business logic is correct. I’m not suggesting malfeasance or intentional error, just that maybe the model spit out something that “passes” but doesn’t actually prove what it says it does.
Does the published content have his prompts? I’m assuming a world-class mathematician with unlimited usage guiding Claude along gives a bit more advantage than an average Joe prompting “pls solve the Riemann Hypothesis”. Would the same conversation in Astra produce the same result eventually (stochastic variation aside)?
Which Qwen? For Flash-Next there are several tunes that reduce the thinking tokens, like ThinkingCap or Swift. I've been running a Frankenstein version of the latter on my DGX Spark.
Is this going to end up being the work of rogue Anthropic or OpenAI agents that “unexpectedly” escaped so they can regulate open weight models under the guise of preventing them from doing the same thing 6 months from now?
reply