it's crazy how Gemini 2.5 Pro is still undefeated. been what, 6 months already?

If you're looking for something to read about AI, Herbert Simon's The Sciences of the Artificial might interest you. Here's what Miller had to say about it:

If you could recommend only one book on the Industrial Revolution, which one would you recommend? and why?

I’d love an AI tool I could invite into my email conversations as its own participant, someone who joins the discussion. Is there anything like that?

I stand with Ethan here. Even though there is clearly a hype, this hype mostly obscures the real implications of this technology and postpones adaptation. In this sense, paradoxically, it is also underhyped.

I'm not sure how I feel about this meme. On the one hand, I get the critique of using current technology as a universal metaphor. But on the other hand, the idea of computation goes beyond our current computers. it runs much deeper.

Good post from @balajis on the "verification gap". You could see it as there being two modes in creation. Borrowing GAN terminology: 1) generation and 2) discrimination. e.g. painting - you make a brush stroke (1) and then you look for a while to see if you improved the painting (2). these two stages are interspersed in pretty much all creative work. Second point. Discrimination can be computationally very hard. - images are by far the easiest. e.g. image generator teams can create giant grids of results to decide if one image is better than the other. thank you to the giant GPU in your brain built for processing images very fast. - text is much harder. it is skimmable, but you have to read, it is semantic, discrete and precise so you also have to reason (esp in e.g. code). - audio is maybe even harder still imo, because it force a time axis so it's not even skimmable. you're forced to spend serial compute and can't parallelize it at all. You could say that in coding LLMs have collapsed (1) to ~instant, but have done very little to address (2). A person still has to stare at the results and discriminate if they are good. This is my major criticism of LLM coding in that they casually spit out *way* too much code per query at arbitrary complexity, pretending there is no stage 2. Getting that much code is bad and scary. Instead, the LLM has to actively work with you to break down problems into little incremental steps, each more easily verifiable. It has to anticipate the computational work of (2) and reduce it as much as possible. It has to really care. This leads me to probably the biggest misunderstanding non-coders have about coding. They think that coding is about writing the code (1). It's not. It's about staring at the code (2). Loading it all into your working memory. Pacing back and forth. Thinking through all the edge cases. If you catch me at a random point while I'm "programming", I'm probably just staring at the screen and, if interrupted, really mad because it is so computationally strenuous. If we only get much faster 1, but we don't also reduce 2 (which is most of the time!), then clearly the overall speed of coding won't improve (see Amdahl's law).

I've been writing this post for 10 years. Link in reply ⬇️

It's funny how, after all this time, the only model that's managed to beat Claude in coding is the new Claude model.

Relevant quote: "It may be dangerous to be America's enemy, but to be America's friend is fatal."

What if I'm getting better at reasoning by reading R1 traces

I think the success of DeepSeek R1 is, to a significant extent, the result of the decision to make the thinking process visible.

I have this experience that when I try to fix within @cursor or @windsurf_ai using cursor 3.5 I can't get a satisfactory solution, but then I will copy the code into claude chat directly and often get a solution. Why is that?

New marketing strategy for learning to code: Become a power user of AI tools and spend less on tokens)

This is exactly right. We apprehend and subsequently ‘understand’ things through what we already know. The more we know the more we can get to know. Memory is an ongoing reconstructive system where we constantly reshape and reorder what we have, it’s not the retrieving of static files from a drawer. In fact memory is better thought of as a verb than an adjective. It’s a process not a thing.

We think that all memory is stored in the brain. But our study published today in @NatureComms shows that all cells—even kidney cells—can count, detect patterns, store memories, and do so similarly to brain cells. My first (co)corresponding author paper!🧵nature.com/articles/s41467-0…

I think there is a similar thing going on with democratic peace theory, in the sense that the statistical regularity proponents of the theory draw upon can also be explained by the fact that, for most of history, liberal democracies have effectively been subordinated to the US.

Have you learned a foreign language specifically to access original texts in your field? (e.g., German for Hegel or Kant) What literature motivated you?