avindroth

It depends a lot on the domain, even for CoT. I don't think there are enough NLU evaluations just yet to robustly compare GPT-3.5 w/ CoT/SC vs. GPT-4 wrt domain.

For instance, with MATH dataset, my own n=500 evaluation showed no difference between GPT-3.5 (w/ and w/o CoT) and GPT-4. I was pretty surprised by that.

I am building an extensive LLM-powered app, and had a chance to compare the two using the API. Empirically, I have found 3.5 to be fairly unusable for the app's use case. How are you evaluating the two models?

I have been mostly using clojurescript for prototyping, any pointers on how to get started with Racket?

CL-PPCRE is great

That's odd, I have been heavily using GPT-4 API (near 100 requests a day) and didn't notice any errors like that. I noticed maybe 1 or 2 errors with really chat history.

Maybe vector-based search? If 1000 words is 1500 tokens, then it’s 3 * 365 * 1500 * $0.20/1000 for embedding with davinci-1. That’s around $300 for full semantic indexing of your journal entries, which is not bad. Of course, you can get away with weaker models to drive that cost down.

Supernotes is wonderful. The closest thing to Zettelkasten imo.

I like FP, but I don't think the situation is solved by FP. There's a lot more room for discussion of abstraction and complexity within FP.

I came into this thread thinking of recommending this book. Are there any other books like it that discuss managing complexity, maybe not just for software engineering?

Reminds me of Bret Victor and hundredrabbits. If Lisp was more popular, we could use it as building block for UI (since it can describe both code data). An email client where all of your emails is a (lazy) list? A Figma-like where you can compose, filter, reduce the graphical elements? Lists are everywhere, and Lisp is built for it.

This is cool; a UI suggestion would be to show what the badges are when you hover, and not click-to-alert.

In Lisp, there is no distinction, and it's a huge advantage.

After years of vim, I started using emacs with an ergonomic keyboard on my tower. Then I tried emacs with my mbp, and it was so much worse (ergonomically) than vim. I think emacs without ergonomic keyboard is just really painful.

To each their own. Minds function differently, desire different workflows, and thus are extended better with different tools.

I use both non-evil emacs and neovim, given the mood I am feeling that day. For quick and dirty things, I prefer neovim, and for more extended workflows, I prefer emacs.

But besides functional differences, the two "mediums" feel different. Neovim is clean (and hangs less in general). No bloat. Emacs is a monster, suited towards my liking (with many years of .emacs config).

But as much as I love Lisp, sometimes you want to get away from the bloat and sit quietly with a blinking cursor in vim.

By a similar line of reasoning, many people did not expect LLMs to achieve these results. Maybe reductionism isn't too useful when we are dealing with emergent behaviors.

The consequence of having the AST be explicit is profound.

Just a guess - you may be downvoted not for exactly what you are saying, but how you are saying it. You may have valid criticism, but being rude is not met with welcome on HN (for good reasons).

Before you comment something like this, ask yourself "How would I make this easier to try?" The only reasonable answer is providing the OP's own API key, which is undesirable.

What’s great about meow’s macro system?

Anything you’d recommend specifically from McLuhan besides Gutenberg Galaxy

+1 to re-frame

CLJS has been surprisingly good. re-frame really simplified a lot of the development. There are some oddities with trying to work out whether something is a JS problem or a CLJS problem, but for a solo project, it has been pretty good. CLJ/S seems to me a good spot between practicality and idealism (CL being the latter for some things). It's the most practical Lisp atm, and it can't get better than that.

I have also been using CLJS/Clojure for a solo project. I debated between CL and Clojure for the backend, but CL’s package ecosystem is too weak to be justified. What other languages did you consider for your stack and what did you think were the tradeoffs?

I really don’t like the argument that you should make things free just because it makes the world better. What happened to ownership and respecting the effort it takes to create something?

I see a similar line of reasoning can be used to justify theft from the rich.

It pretty much was.

I think it's a balancing act between labeling and mislabeling. Some things are so hard to name that naming them actually hurts. Also sometimes you don't want to pin down what it actually is; naming restricts the concept.

Wrt maths, you did maths without greek letters and with plain english, you'd have pretty ugly formulas. Greek letters, while obtrusive to newcomers, can be a lifesaver for conciseness in expression.

Thanks, I will fix that.

Ironically this page is completely unreadable on the sidebar of my iPad, making the text ugly.

As a user of Common Lisp but not Clojure, where does Clojure dev environment stand with respect to something like SLIME? What unique features do Clojure dev environments have?

Is there a reason why some well-known Lisp hackers are Japanese? I noticed the devs of Kandria (radiance, etc.) as well as Fukamachi (woo, clack, mito, etc.), and am wondering what led to that kind of geographical distribution.