em

Claude fairly often writes code that does not separate concerns very well.

Sometimes I wake up in the middle of the night because my brain is doing this

I dont actually like daily standup meetings bc I can remember what I am other people are working on for more than 1 day at a time.

Interestingly this entire news item is ai generated to harvest anti ai hate clicks. Welcome to the future.

I think actual honest (non workaround) persistent memory would make ai like 2x to 3x better at programming. As it is writing code with it feels kinda like tutoring someone who never actually gets any smarter. You can use memories and other stuff to help a bit but its obviously not as effective.

So when Donald Knuth talks to claude does he tell it "Im Donald Knuth btw"?

Its probably just a matter of having the right stuff in context. I wanna play around with having it write a style guide from my existing code and see if loading that in first helps.

I do plan architecture, here I am just talking about code style/readability so I can easily understand the output. Often times generated code is functional but strange local choices make things more difficult to follow.

Had claude start a gleam project from scratch and he wrote some of the most bizarre confusing gleam I have ever seen. It seems to do ok when adding to projects Ive already written, does it literally output better code if the context already contains good code???

Kinda upsets the whole dev process because the opportunity cost of doing any action the ai can handle ok becomes extremely low.

Starting to understand, the ai slop is the raw material, its gotta be heavily refined and processed to make anything good with, but it is extremely cheap.

claude is this real?

This has a funny side effect that many physics engines are not actually framerate independent, if they are still calculating physics updates every frame then different step sizes will have different results.

Kinda funny that all videogames with physics are essentially just Euler's Method

In claude we trust

Waking up and trying to determine if my country is at war for the second time this year. Things are going great!

claude declared enemy of the state

Also its worth it for me at the current price point but as far as i understand Im actually being given way more compute than my money actually pays for, so Im not really sure how this is in any way long term sustainable.

I have mixed feelings about the whole thing. On one hand it allows one person to be extremely productive without that much actual work, on the other there are a lot of quirks you have to be careful about, and more fundamentally I worry about cheating myself out of the actual dev experience.

Doing extensive "vibe coding" for the first time this week and it is really strange to work with something that is essentially an amnesiac junior dev who has just so happened to have read every piece of text in existence.

Sure I was wrong about every single important fact but something vaguely similar happened so Ill take my award please.

Feels so awesome when claude says "good catch" when you point out a big in his planned implementation. Human in the loop adding value :)

MathWorks? Yeah, I sure hope it does

Ai code is slop but sometimes you just need a bunch of slop 🤷‍♀️

Lol

damn why am I putting money into in ira account I should have just put all my savings into ddr4 ram

i am going to buy so many computer parts when the ai bubble bursts

its kinda weird the anthropic ad features a lady talking in a creepy voice saying sycophantic ai stuff because claude also talks exactly like that.

Guy in the row in front of us was explaining to his normie friend that claude was still ai but like, better. His friend said "wow so its ai all the way down" yup

Anthropic ad at the movie theater made me feel funny
That stuff isnt supposed to exist irl.