tipsytoad

I wholly disagree with the comic, but a anti AI art take I’m more sympathetic to: https://x.com/soi/status/1815584824033177606?s=46

I don’t think this hits at the heart of the issue? Even if we can catch AI text with 100% accuracy, any halfway decent student can rewrite it from scratch using o1s ideas in lieu of actually learning.

This is waay more common and just impossible to catch. The only students caught here are those that put no effort in at all

Just another +1 that if you’re going to give vscode a fair shot, it’s much better to go with vscode-neovim than the standard vim extension. You can even map most of your config right over.

E.g. (mine) https://github.com/tom-pollak/dotfiles/tree/master/nvim

How is this different from instructor? github.com/jxnl/instructor

namely, why did they take so long for something that just seems like a wrapper around function calling?

Looks really nice, but the only concern I had — how does the perf compare to more mature libraries like faiss?

The most useful feature of llms is how much output you get from such little signal. Just yesterday I created a fairly advanced script from my phone on the bus ride home with chatgpt which was an absolute pleasure. I think multi-prompt conversations don't get nearly as much attention as they should in llm evaluations.

Switched to Firefox a month or two ago, mostly for ublock origin on android, and unlimited history (seriously, why is this not the standard?)

I've tried to like it but honestly it's been painful. MacOS Sonoma seems to have a hover bug, which has been unresolved through the last 3 bug fix updates. Performance is "fine" but seems to lag with many tabs open which was never an issue in chrome (this is on an M2 pro!) PDF reader also seems significantly slower as well. At this point I'm considering going back to chrome.

Moved to vscode with the neovim extension (vim mode was slow for me over large files). vsc is super customisable, and you can remove all the tab bars, side panels and anything else you don't use fairly easily. It's also far more stable than the neovim ecosystem, i don't have time to mess with neovim plugins breaking bi-weekly anymore

Using fzf through bash history and using # comments as tags on common commands seems to work just as well for me. Not sure what else this adds

Eh, you can probably get very similar to fish with zsh + loads of plugins, but fish has lots of niceties out of the box (syntax highlighting, autosuggestion based on your directory). I've been using fish + starship with just a fzf plugin and it's got everything I need.

Check it out, it's effective with very little config

Priding ;)

So like an autoencoder for text?

I emphasize with the student here, but passing it through an ai detector yourself before submitting and keeping a version history seems like an easy way to combat any false claims.

It's a repository viewer last time I checked, it doesn't have as many features as lazygit (rebase, merge conflict). I prefer lazygit personally

Tangentially related, but if you don't have access to c++23 or you don't like monadic properties, passing by reference and returning bool success can emulate or_else, transform, and_then with || &&

  bool fetchFromCache(int userId, UserProfile& profile);
  bool fetchFromServer(int userId, UserProfile& profile);

  UserProfile profile;
  if (!fetchFromCache(userId, profile) && !fetchFromServer(userId, profile)) {
    std::cout << "Failed to fetch user profile.\n";
    return;
  }

or just to prove a point, although even more unreadable than the monads:

  bool extractAge(const UserProfile& profile, int& age);

  UserProfile profile;
  int age;
  if (
    (!fetchFromCache(userId, profile) && !fetchFromServer(userId, profile))
    || !extractAge(profile, age)
  ) {
    std::cout << "Failed to determine user's age.\n";
    return;
  }
  
  int ageNext = age + 1;

This probably isn't a great example usecase for this, but its helpful when creating composable transform pipelines

Karabiner and applescripts can get you pretty far

Python, C++

Really? The only reason that ChatGPT is more adept at coding problems is because there is vastly more training data. There's nothing fundamentally different between problem solving a coding problem and physics problem. Like all the others before it, I don't think this comment will age too well.

I like https://github.com/PatrickF1/fzf.fish

Can also do cool stuff like search for a file to open in $EDITOR and searching git log

Seriously, are they only talking about weight decay? Why so complicated?

Your CV is how many pages? Instead of listing everything from assembler to JSON, maybe try to get it down to 1 (2 max)

Lol. Have you ever tried to use a trackpad on Linux?

Vscode has this built in: https://code.visualstudio.com/docs/python/jupyter-support-py

I didn't know about Protocol! How does this differ from ABC?

Materialistic is a good android app for HN

Trackpads SOTA, Keyboards OK, Mouse is PITA

Boy I can't wait to remap my entire vim/vscode config, relearn a lifetime of touch typing, and have my computer inaccessible to anyone else who wants to use it for a +30% typing efficiency

I agree? Also related karpathy:

https://karpathy.github.io/2021/03/27/forward-pass/

There's no "exploration" here... The argument that an LLMs output can be deterministically calculated as a sum of matrix multiplications and non linearity is not evidence that LLMs are conscious/understanding either way.

In reference that LLMs "don't understand anything and are stochastic parrots over their training dataset" refer to in context learning, llms can be taught new concepts not present in their training data.

Dude.

Cos(x, y) = x . y / ||x|| * ||y||

If x and y are normalized:

Cos(x, y) = x . y