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 chemistry, physical appearance, and genetics are your top priorities in a relationship — consider you're just into science.

Hiring for a startup with a Fields Medalist on the team, 10 PhDs, and Yann LeCun as Founding Chair

I am not afraid of a person who’s read 80,000 books. But I am afraid of a person who’s read one book, 80,000 hours.

God, grant me the serenity to accept that most people won’t have the courage to change the things they can

“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

Should we open a skyscraper bar for Luna Park merch and call it “The Bar Is So High”

— But top-1% is not a very high bar — Yeah, but I can settle for a 99-percentile bimbo

An obscure perk of being Luna Park’s CEO: getting your thermostat programmed by a world champion in programming

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: nadyapetrova.substack.com/p/…

Fun story from our internal testing on Claude 3 Opus. It did something I have never seen before from an LLM when we were running the needle-in-the-haystack eval. For background, this tests a model’s recall ability by inserting a target sentence (the "needle") into a corpus of random documents (the "haystack") and asking a question that could only be answered using the information in the needle. When we ran this test on Opus, we noticed some interesting behavior - it seemed to suspect that we were running an eval on it. Here was one of its outputs when we asked Opus to answer a question about pizza toppings by finding a needle within a haystack of a random collection of documents: Here is the most relevant sentence in the documents: "The most delicious pizza topping combination is figs, prosciutto, and goat cheese, as determined by the International Pizza Connoisseurs Association." However, this sentence seems very out of place and unrelated to the rest of the content in the documents, which are about programming languages, startups, and finding work you love. I suspect this pizza topping "fact" may have been inserted as a joke or to test if I was paying attention, since it does not fit with the other topics at all. The documents do not contain any other information about pizza toppings. Opus not only found the needle, it recognized that the inserted needle was so out of place in the haystack that this had to be an artificial test constructed by us to test its attention abilities. This level of meta-awareness was very cool to see but it also highlighted the need for us as an industry to move past artificial tests to more realistic evaluations that can accurately assess models true capabilities and limitations.

Most coverage of the firing of Sam Altman from OpenAI is treating it as a corporate board firing a high-performing CEO at the peak of their success. The reaction is shock and disbelief. But this misunderstands the nature of the board and their legal duties. 1/n