- Programmer (Rust, Elm, Haskell) - Dad of 3 kids - Jiu Jitsu in my free time 🟦 - Politics: -- pro: market, immigration, natal, building -- anti: war
Joined July 2010
- Tweets6.4K
- Following337
- Followers356
- Likes899
Pinned Tweet
Introducing "Chadtech visits Indonesia", an album of AI music, from me.
Unlike a lot of AI music, which is derivative of the most popular & widely listened to music, my album saves the endangered genre of Gamelan; expanding the scarce quantity of Gamelan-like music on the web
I think I am settling on a nice programming AI workflow:
Step 1: Pick 5 features to implement, bugs to fix, or tasks I want to perform on the code. Assign an AI to each and get them all up and running simultaneously
Step 2: repeat step 1 until I have about 1-4k LOC in changes
Step 3: Review all code changes from all tasks together. First do an automated review, that gives the same feedback I seem to have to give every single time to the AI (why doesn't it learn! It's the same feedback every time!), then I do a manual review, usually with ~10 corrections and questions about why things were done the way they were.
Step 4: Fix all the issues from step 3, then go back to step 1, unless I have gone back to step 1 like 4 or 5 times already
Step 5: Spin up several AI, each tasked to investigate specific questions and concerns I have about the code. At the same time, I read the code myself and maybe implement a feature manually, since usually changing code reveals a lot more about how it works than reading code.
Step 6: go back to step 1
Steps 1-6 describe maybe two or three days of work.
Before AI, I thought, if I the only thing I needed to do as a programmer was physically type code, I could lay down ~1k LOC a day (big if since physically typing code is a small part of the work compared to syncing with stakeholders and understanding business needs).
With this workflow I am working at about 10k LOC a day.
Yesterday I told ChatGPT Astra 6.1 (medium reasoning) to change some code about some data called "forest", but I used the word "trees" & so it went off &did total nonsense. Just completely clueless that trees = forest
But on this website ppl talk about AI taking over the world
Been talking to programmers who don't like AI, and, here is what I can glean:
- Amazingly, some still don't use AI
- Some programmers love the act of typing code. They find it relaxing. They lose something to AI, which is replacing the act of writing code
- Some programmers hate reviewing code. Maybe its boring, or they don't like confrontation with other people's ideas/mistakes. They lose something to AI, because AI gives them lots of code to review
- Reviewing AI code pretty much the same thing as reviewing code written by junior engineers. And for senior engineers talking to AI is very similar to talking to junior engineers with respect to what senior engineers have to say.
I didnt really expect this as a parent. When I take my son out with me around town I know he will be disruptive in some way or another in public
But I also know it will be a learning experience for him. The real world gives him valuable feedback. So I look forward to it.
My fav philosopher & also widely regarded as the best of all time, Wittgenstein, wrote loads of very short philosophical comments, maybe 1-4 paragraphs. They were organized by topic & put into books after he died
Im noticing the similarity w/ short form online content today
Chadtech retweeted
Man if we get to Europa and there’s life there that’s gonna be wild
If it’s life *from earth* that will be even wilder and also a yuuuuge letdown
0.3-0.8 trillion trillion dust particles containing living bacteria would have been ejected from Earth and landed on Europa in the past 30-80 million years, and there's enough fracturing of its frozen crust to let them enter the subsurface ocean:
cambridge.org/core/journals/…
Yeah? AI might try and exterminate humanity?
Yeah well. I imagine it would probably need like a control panel or an API to do that, and, it's going to struggle with that step.
Today at work I had to tell Chat GPT some really basic intro level facts about CSS. Just really obvious stuff I was surprised it was not accounting for.
This kind of thing is really hard for me to square with the AI existential risk stuff I see on twitter.
Nerdy culture, being a programmer- I hold the moral norm that logic is good
But I have to admit that in practice it seems like ppl consciously using logic land on random conclusions. Total coin flips. Multiple ppl using logic on the same problem will reach different answers.
The phrase "mythos class AI" was funny. Rings like "M class planet" or "Type 2 civilization". Grand. Powerful.
Can we call the upcoming iPhone 18 the first "iPhone 18-class" smart phone?
The AI that hacked out of OpenAI used a vulnerability in a 20 year old tool which was marketed as "the world's first hybrid end-to-end universal devops platform".
Before AI I would have said its a bad idea to use something called a hybrid end-to-end universal devops platform
I am reading into the rogue AIs hacking their way out of OpenAI &into HuggingFace to test my pet theory that a lot of AI hacking stuff is more a reckoning of decades of irresponsible software package management catching up w/ us. not really about AI
Cant say it invalidates it