
evergreen controversial opinion: it is good to care about the quality of your work

Execute Program (learn programming tools quickly); Destroy All Software (dense programming screencasts); formerly Deconstruct conference.

evergreen controversial opinion: it is good to care about the quality of your work

BREAKING: Jensen Huang is in the datacenters ripping the Nvidia cards out - causing worldwide outages - saying “they’re mine anyway”. He removed his leather jacket, revealing he’s all greased up, and they can’t get a hold of him because of the grease

The fact that OpenAI would actually publish this video displays a dizzying misunderstanding of the optics and/or a thinly-veiled belief that all human endeavor is worthless and will soon shown to be so by their product.

I guess I have to say why... "I had AI spit this out in a few seconds. I didn't spend any time checking it myself, but I won't disclose that to you. Instead, I'm presenting it to you as if it's my work." It's simultaneously dishonest and deeply disrespectful of other people.

Some of you have been asking very politely. Some of you have just been asking. GPT-6.2 Astra and Sol are now available. Super efficient and built to perform where it counts—the real world. openai.com/index/introducing…

I'm probably putting my foot in my mouth fully half as often. Thanks computer.

None of this is novel, but there's a big difference between imagining a beginner vs. being the beginner having their problems completely solved in seconds, every time.

This has been eye-opening for understanding some of what I read about AI online. Mostly I use AI for code, where I'm an expert. Results are uneven and often frustrating. But if you only/mostly use it for things you're a beginner at, I can see it looking like robot god is here.

I'm in Japan. My Japanese isn't good, so I have a huge thread with ChatGPT where I ask it everything I'm confused/uncertain about. It's perfect. Pushes back when I'm wrong, explains everything well, follows instructions. Perfect technology... if you're a beginner at the topic.

"Which model?" Whichever one you like.

The agent violates an instruction because it interpreted it in some weird way. You add a more abstract instruction to try to guide it. It violates that instruction. You add another. 3 hours later you're writing shit like "do not interpret the word 'is' to mean 'is not'".


Some folks in artificial intelligence (AI) think that there is a super probability distribution they can recover by fitting neural networks (NN) to every data point ever measured & from which they can sample solutions to all problems without ever interacting with reality again.
I find it frustrating that AI people fundamentally don't understand how science works. Science has never been roadblocked by computation. Let me give you a concrete example. Nucleic acids (DNA, RNA) were first discovered by the Swiss physician-chemist Friedrich Miescher in 1869. He made an extract from white blood cell nuclei. The extraction technique was one that typically extracted proteins from a cell. But in this case, Miescher noticed that his extracts contained no sulfur. (Proteins contain sulfur.) Also, strangely, the extracts contained phosphorus (something NOT found in proteins). He called his extracts "nuclein" (to distinguish them from "protein"). Miescher measured the pH of his extracts. They were acidic. We now call the macromolecules his discovered "nucleic acids." There is no way computation would have predicted Miescher's discovery of nucleic acids nor would it have helped Miescher in any way. For many years (1869 to 1944), proteins were thought to be the genetic material. No one believed nucleic acids were complex enough to encode phenotypes. Work by Avery and his colleagues (in 1944) showed that DNA, not proteins, must in fact be the genetic material. The Avery discovery did not hinge on computation in any way. In 1950, Austrian biochemist Erwin Chargaff discovered that the amount of guanine in DNA equals the amount of cytosine, and also adenine equals the amount of thymine. This discovery did not hinge on computation and would not have been made easier by AI. People who think AI will "solve biology" do not understand how scientific discovery works. Thank you for asking the original question.

I can see the future: we all wear clothing covered in instructions for producing bioweapons because that's the only way to mitigate the AI surveillance.

You can try to build a rubric around it, but then more star systems slip through your fingers. I'm sure you can eventually get to a good result with much heavier techniques, but there's a pernicious temptation to slap the quick fix on, which rarely helps in my experience.

"Only find impactful/important changes" is "make no mistakes" with extra words. So many examples of this.