Andreas Møller
Top 99% software engineer.
Girl dad and boy boss.
Co-founder of https://nordcraft.com
- [Not loaded yet]
- You are leaving out all the relevant information. How many operators do they have? How often do they have to intervene? No body is saying AI cannot drive a car, the question is can it drive autonomously.
- Too many good people in tech continue to platform the obviously bad people in tech because of their success, their audience, their notoriety, etc. Holding yourself accountable to who you choose to platform is a sign of self-respect. Otherwise you're just another weak opportunist.
- [Not loaded yet]
- Is his firstname almost felon?
- I don't know if AI makes you 10x faster, but LLMs are definitely "10x developers" THE TESTS WERE ALL GREEN WHEN YOU STARTED!
- border-shape can handle both insets and outsets, so you can do effects like this chevron nav (corner-shape can't do both). This means you get a perfectly-wrapping focus ring without needing to manage z-index or having it partially covered due to overlap. Demo: codepen.io/una/pen/ByzY...
- Very happy about this!!! 90% or the time I am using clip-path, I am trying to emulate this
- In my experience code quality has never been more important than when using AI agents. If I stop supervising the LLM the code deteriorates fast and the Agents error rate skyrockets. It is not just about concise code, It is just as much about structure.
- I've noticed the same. The agents I work with perform noticeably better on well-structured codebases. Do you find it's more about file/module boundaries or naming conventions?
- Structure is definitely important, but bad naming can definitely trick the agents up as well
- Here’s one that’s not going to happen
-
View full threadyes, but their capability of understanding the present would be limited by their knowledge cutoff date though they could still use web search, as long as the web also doesn't turn to unusable slop (highly likely) i wonder what longitudinal research will reveal
- This.
- [Not loaded yet]
- “Why do I need to learn to read when chatGPT can?”
- Nothing ever moves in straight lines, but the idea of models getting worse is definitely a possibility. We had a very brief view of that when GPT 5 came out. As the amount of AI generated content grows, It will be harder to separate quality training data from slop.
- I was going to write a whole effortpost about model evaluations but instead here, just take this chart
- I think it is a very flawed assumption that the green line stays flat
- [Not loaded yet]
- Yes, people who cannot use a mouse are not going to learn if we keep cuddling them.
