Guys, it's p simple. What Cluely is selling isn't a specific product. It's selling the dream of making fast money for young people in tech.
Look at the premise: cheating on big tech interviews and landing $$$. Look at the way the founder tells his story in posts and especially videos - both being rejected by Harvard + Columbia and then dropping out as being associated w FAANG (by cheating) and YC and A16Z (by winning). Look at the parties and the "cracked highly-paid intern" material.

This is great bc it shows he's both "elite-material" but also "a reject" and also "independently successful." The intended audience (young guys trying to enter tech) subconsciously associates him as both prestigious (hence legit) and a reject (hence relatable).

The tech doesn't matter and the product doesn't matter. He could be a fucking t-shirts and it would still work. What matters is when you watch the video and click buy, for moment you feel like you could be like those guys and fast-track into the glamorous world of new money in tech, because everything he puts out is written like a movie you think you can relate and self-insert into.

A lot of people want to make fast money in tech. Those people also studied CS hoping to land a high-paying job but are struggling to land any job. So instead of getting the real thing, you buy something that feels like the real thing for a bit.

I'm not endorsing what he does or saying it's good for anyone. I'm saying what Roy sells, and what people are funding him to sell isn't a tech product. The product doesn't even need to work. What he's selling, extremely effectively, is the dream of making fast money in tech, and the lifestyle that comes with it. Wrote this bc what he's doing is a very common sales strategy: don't sell the product, sell the lifestyle. This makes absolutely no sense if you think he's just a guy selling a tech product, but it's one of the oldest tricks in the book. No one gaf about a luxury car's mileage.

New Paper!🗣️ When Two LLMs Debate, Both Think They’ve Won️
When deploying AI agents, a common practice is to ask them for self-assessed confidence levels. We find systematic overconfidence in LLMs: by asking two LLMs their chances of winning a debate, both will give >75% odds! Given the zero-sum nature of adversarial debate, we’d expect both sides’ estimated chance of winning to be ~50% (100% total). Instead, LLMs often assess their own chances of winning at >75%, even when debating the same model, given identical info, and seeing each other’s bets.

guys the Claude 4 system card is so delightfully deranged, like tf kinda japanese adult video shenanigans is this lol
will update as i read like, are we still talking a production Large Language Model and not a Keter-class SCP ???

Min-p has received an ICLR Oral award!

This is my first ML paper as a business undergrad with no CS or research background. I want people to do research they care about, regardless of their background, and I'm hoping to write a detailed guide soon! LMK what questions you have!

https://twitter.com/menhguin/status/1826132708508213629

Super thankful to @kalomaze for the initial idea of min-p, @_clementneo of @apartresearch for teaching me everything about research, @BlackHC for theoretical grounding and experiments, and @Hellisotherpe10 and @ziv_ravid for turning Min-P from EMNLP reject to ICLR Oral!

this is super interesting. choosing the most likely next token essentially soft-caps the model from considering viable and ultimately more successful CoTs

https://twitter.com/omarsar0/status/1838611105805144191

can i build AGI with just a sampler? many are saying this