No joke not dunking on you: try this. You'll learn something and programming is fun. Harvard's CS50 or CS50p course might be a good start - it's online, free, and pretty good. cs50.harvard.edu/x/
There are no specific "finding errors in code" books what you'd need to do is learn to program.
Look at the conversation I just linked and tell me with a straight face that it's possible to solve that problem without thinking.
They obviously do? How do you imagine they write code or solve complex problems if they can't reason? I got one to do abductive reasoning here, for example bsky.app/profile/did:... . This was the first time I tried to do this; it's not cherry-picked. How do you think it can do this task?
www.anthropic.com/research/int...
(Also obviously they have internal representations how do you think they think?)
I think the people who say Claude code isn't impressive because programming is easy might actually believe this
Also human translation isn't 100% reliable either and the benchmarks are in favour of the models here.
That's fair, although to be more pedantic many problems in P are easier to verify too; e.g. you can check if a list is sorted in O(n) and sorting it will take O(n log n).
But also in this case it's obviously easier to verify a small chunk of a language you're not great at than an entire document.
In addition to the like react bluesky should have a laugh react, for when someone has made a good joke or made themselves into a good joke
You might want to let the computer scientists know about your breakthrough in the P=NP problem then!
Livesplit display showing a new best time of 27:20.35 for Windswept Any%
New personal best for Windswept Any%
"the person I am making fun of specifically and also the entire group they are metonymy for did not do the thing I am making fun of them for, and in fact did the exact opposite" is not a small exception or edge case.
When I visited the US on a holiday in the mid 2010s it did not seem especially wealthy to me. In many ways it seemed crumbling - public infrastructure was not good.
Their writing on data centres is also extremely misleading: andymasley.substack.com/p/more-perfe...
Strictly speaking I think it's China Brain, but there's this whole interconnected set of Thought Experiments That Differentiate Functionalists And Others
Before LLMs my very small workplace hired a middle aged guy who'd been working as a programmer for quite a while who we quickly realised couldn't write a line of code. Sat on an interview for a grad at the same place who also didn't seem to know what he was talking about.
There's absolutely still Guys Who Think It Can't Code. Also people who just generally hate it for all the usual dumb reasons.
The Wachowskis: In our films we invented the Bullying the Robots as a cautionary tale
Leftists: At long last, we have created the Bully the Robots from classic sci-fi film series Don't Bully the Robots
It’s important to be precise with your language, people!
Mr. Darcy never actually *say* or *feels* anything, because those words describe people, and Mr. Darcy is just words on a page, and words have no animus of their own.
Surely it's more likely they were bullies in school
I've used Claude Code a bunch for writing a livesplit autosplitter for a game which people are definitely using (also I'm using it and it's handy). Also unit tests for a mod for a digital card game, also used by people. But I'm not hitting token limits meaningfully.
I don't really see it for Shaun of the Dead; can you elaborate?
And I'd argue that World's End's rejection of regulatory busybodies telling you to do things for your own good is also pretty liberal!
A cute Calico cat sitting on a pile of board games looking at the camera
Let me help you with that! This is Chloe
This brought to you by seeing Moulin Rouge with my partner for Valentine's day.
Story-within-a-story is a banger trope. So many things it can drive - dramatic irony, characters talking to each other in character but actually making it for real, general reflections of main story in sub story, confusion of meta levels. I don't think I'm ever disappointed to see it.
This story is from 2018
Have you just not heard of Claude Code and related products? It's an incredibly useful productivity tool for programming.
(Also you're undervaluing "do some web searches and summarise the results for this goal"; it's useful to be able to say "go find me a bluesky client with these properties")
End of History 2: Electric Boogaloo
It's incredible how fast people decide that writing a gcc-compatible C compiler isn't very impressive because ha ha the default include paths are bad or three of the documentation tests fail or some similar bullshit.
Do you know if someone's analysed the code to work out why it's got worse codegen than gcc with no optimisations? Maybe I'm underestimating how much gcc does at O0 but that strikes me as suggesting the models have made some kind of serious mistake and the various passes just don't work.
