I run Luna Park—we hire S-tier tech talent (like IMO/IOI winners) for sexy startups and AI safety labs. Besides, I think and blog about psychology and ethics
I recently spoke with a CEO of a crypto company. He said they make complex offers to candidates: salary + token offering. I asked if he called it complex because it has both real and imaginary parts, but he didn't get it :(
If we already have AGI — why can't it solve a sudoku puzzle?
According to any reasonable definition, a "reasoning" AI should be able to arrange a few numbers in a 9x9 grid without much trouble. But frontier LLMs are really bad at this.
By "this" we mean Constraint Satisfaction Problems — stuff like chip design, energy grid optimization, HFT. This is because token-by-token generation tends to commit to bad choices very early on, and can't revise them later.
However, there is a whole other approach — Energy-Based Models, and researchers like @YannLeCun have been pushing for it since the 1980s. Now it finally seems to be feasible, and, among other things, can give us cheap and efficient formal software verification — which is very much needed, given just how messy ALL of humanity's software turned out to be.
EBMs minimize an energy function in latent space. High energy means high constraint violation (something's wrong); low — you're close to the truth. Unlike LLMs, they optimize the entire trace at once — and can refine it iteratively. This allows for much more precision with much less compute 🧵
“Some people are making us believe that we're really close to AGI. We're actually very far from it. I mean, when I say very far, it's… several years.” —@ylecun
Imagine you recruit the top 0.01% of engineers and care about AI x-risk.
Would you work with AI SWE agent startups (Cursor, Devin, Factory)?
Tradeoff:
– Small direct harm
– Stronger second-order benefit: prestige grows funnel → more chances to steer top talent to AI safety work
@MostlyMonkey tldr: Having two Logical Reasoning sections doesn't sound to me like they don't filter for high IQ at all :) And I'd imagine that an IMO winner with an IQ of 160 but a verbal score of 100 wouldn’t make a better lawyer than someone with an IQ of 130 and a verbal score of 130
@MostlyMonkey I mostly agree, but to be fair, when we make a test less g-loaded, we make it more something-else-loaded. And for most domains, you can benefit by exchanging an IQ point for a point of other helpful quality — especially if you’re already high-IQ
@MostlyMonkey Points of IQ (or any other quality) have diminishing marginal value after a certain threshold. So I won’t be surprised if a perfect test for lawyers is less g-loaded than before, bc it sets a higher bar for something else while also maintaining a fairly high IQ threshold
Like, we grew by word of mouth among tech unicorns, a16z and Khosla-backed startups etc — and I keep hearing “wow, you really understand our needs and our bar! We don’t usually expect that from recruiters” every so often. And all it took was some domain knowledge? Oh my.
We are organizing a Competitive Programming & Olympiad Math meetup in London on July 13th! A bunch of ICPC and IMO folks are already joining so if you're into these things, welcome!
Also, please share this with your London friends who might be interested :)
Finally launched my Substack — time to start blogging in English! Here’s my first piece, discussing how people form their intuitions about what love is: https://t.co/2uch6czoNF
@RafaRuizdeLira None of my purses were as popular as this one today! :)
(And honestly, it’s sort of a bar — I have a purse in a shape of a condom and a jam jar…)
@Maximum_skull I also wouldn’t expect kink-based attraction to be narrower! Many kinks can be triggered easier and in a wider variety of situations compared to how people experience “regular” sexual desire
I have a theory that the nature of attraction to certain gender can be either orientation or kink.
There are kinks for authority figures, old people, or doctors — it’d be really odd if there were no kinks with such property as gender.