AI bros had totally heard of Lean before two seconds ago, how dare you
@futurebird not sure they're clear on the distinction, no
@futurebird @davidgerard I’m more open to trying the former too, I think it’d be healthier
There is nothing wrong with lean. It's a great project. It's not intrinsically linked to AI none of the big computer theorem projects have ever been since the goal of automating theorem proving and checking is as old as the transistor... nay, it may been even older.
(Very NB this isn't my area so what follows is pure ill informed conjecture!)
LEAN is a very useful tool but the problem is it can verrify a proof thats 300 pages of stuff absolutely incomprehensible but checks out logically as correct. So nothing useful just a lot more work for human mathematicians...Doctorow's reverse centaur...but thats not the *fault* of LEAN
@futurebird @at @davidgerard To add to this: any improvements automating theorem proving are tangible steps forward to truly secure software.
@futurebird the Lean team have definitely been slopping it up, though by now none of the big proof assistants like Rocq and Agda are clean (Agda had to be forked into Mikan because they’ve been slopping in new and obvious soundness errors, of which there were too many in proof assistants to begin with), and despite all their bluster it’s a definite step back when it comes to the things that drew me to proof formalization in the first place (e.g. you can’t do homotopy type theory in it)
Purple drank sounds like something made up by an LLM.
Purple drank is just a 21st century version of a Gin and Tonic.
@davidgerard I am trying to figure out how to set up a consultancy to sell devops teams everywhere on this revolutionary new development thing called the Waterfall Model. (Then after two to three months I'll pivot to Jackson Structured Programming.) If the babies have never heard of the greybeard techniques and the web sources have turned to sewage everything old must be new again, right?
@cstross just figure out a way for it to require ever growing amounts of energy and you've got the Saudi funding guaranteed @davidgerard
@ehproque @davidgerard A bit late for that, surely? China's renewables policy is chainsawing away the branch the old bits of the Saudi economy are sitting on. (Saudi policy is to diversify their sovereign wealth fund, hence crypto/AI, but eventually the Coyote is going to have to look down …)
@davidgerard short term thinking; it's not like energy demand is dropping @cstross
@davidgerard It's called that because it lets you express the mathematics more tersely right? Yet Anthropics' Riemann solution is somehow more lines of code than Chrome.
At that size I feel like it's as likely to have found a bug in Lean as solved the problem.
Anthropic will be happy either way of course - the news cycle will have passed on long before anyone can prove it's nonsense.
@glenatron it's given lean more robust testing than it's probably ever had
@davidgerard got something to do with "The Lean Startup", right? i assume so because it's also something none of them have ever actually read but also about which they will not shut up the fuck.
@davidgerard
You mean like when they mix half a bottle of cough syrup with vodka?
Or the maths engine?
The former seems more of the moment.