Daniel Droke

Tried training a language model from scratch with pure RL today Everything became kawasaki

Nvidia is 6-8% of the S&P500

There is a really strong similarity between dogs and beaches in text embedding models

The UX of software is plummeting. I take a lot of pride in trying new software products. I try something new every single day. But I'm finding that there's a huge gap between what I want from a product and what the product wants from me. Like I click a button expecting one thing, and I'm immediately given another. Eventually, I just get pissed off and delete it. So yes, you can write a million lines of production level code and I still won't use it. If you're going to put something out into the world, make it elegant. Build something small and beautiful.

Keep an eye out for CRWV earnings today

People are way too obsessed with LLMs. LLMs are by all means "intelligent". It's amazing that I can talk to a model in natural language. But the idea that language is all that's needed to produce human-level intelligence discredits the human mind. Language is a collapsed representation of thought. It's one output method of the minds ability to express itself. Just because we can build a model to predict one output method of the brain, doesn't mean that this model is naturally equivalent to the human brain. You can argue that all we need for digital tasks are LLMs. But don't pretend that LLMs are the end-all-be-all of AI just because it can write essays for you about hard concepts. Also, the idea that the science of AI is finished is a limiting belief. You are holding yourself back by assuming labs have already solved the "final problem".

I'm starting to think attention might not be all we need...

Are people actually buying things with AI?

I don’t think this is “the pop”. Nothing particularly happened. Probably more like a warning shot. I hope we are realizing that indexes like the S&P are not particularly “diversified” if most of their stocks are AI.

Inference is about to get really expensive.

The AI doomers have rebranded as AI bubble doomers. First off, it's not a bubble. You can't see bubbles. Anyone can see Coreweave's debt-to-equity ratio. It's pretty clear that there will be some market reset in the coming years. Second, you just spent the last few years speaking out about how AI was too powerful and going to replace everyone. That there was no more value in intelligence. But NOW you believe that AI can't produce enough value for society to justify the spending. Either you were wrong then or you're wrong now. Which one is it?

I've been having too much fun having claude generate jupyter notebooks for me to run through

On October 29, 1929, the U.S stock market crashed causing the worst economic meltdown in US history. Yesterday, October 29, 2026, exactly 97 years later, Google lost 12.6% in one day marking its worst day in history. The market continues to fall. Over $2 trillion gone in the past two days. History might not repeat itself. But it certainly rhymes.

Crazy that Doordash bought Instacart for half of Instinct

Why is Jensen always wearing a black leather jacket?

Wispr Flow needs to drop a mask like this so I can talk to my agents in complete silence

Crazy how fast we went from “AI is taking over” to “the AI bubble is about to pop”