geek of programming languages, operating systems, and hypermedia platforms
Joined April 2008
- Tweets5.4K
- Following1.6K
- Followers5K
- Likes14.1K
LLMs are incredible at producing something that appears to work, but if you look at in detail, it's a series of Rube Goldberg machines.
I don't believe it's a skill issue on my part. They will invariably do something incredibly stupid for the tiniest challenges. I still love them, but they certainly can't program.
Thanks, building a cool humane internet is 1,000,000x more interesting than agent shit.
Yes, the internet if it was true hypertext would look like Obsidian if it was networked knowledge graph, not a personal one. The internet with a complete interest graph would look like Goodreads if it covered all canonical work, not just books. The internet with a universal typed ontology would look like Wikidata if it was self-governing, not editorial. The internet with a complete content graph would look like Pinterest if it let you curate all kinds of content—articles, websites, etc.; even users, other boards, chats, video— not just images. The internet with a complete graph would look like Twitter if sharing something was deduping it to discover it's existing commentary and connections, of anything online, and join that discourse; discover chats that engage it, and let discovery be colored by your own personal relations, not a general search but one relative to the shared taste of your peers.
Lisp syntax in particular and syntactic metaprogramming in general come with various trade-offs. It's perfectly understandable if people don't find them applicable to their use cases.
That said, Lisp's procedural macro system is an absolute pearl of PLT.
The way you can immediately use a macro once it's defined, the way macros can generate new macros, the way you can load arbitrary utility code at compile time for use by macros - all of those make this a dream environment for metaprogramming.
(It is tricky to implement though. You either need to be able to dynamically link macro code, or an additional interpreter. This is why we see stuff like non-procedural macro systems (syntax-rules).)
I fully endorse browser maximalism.
This year, millions of people who never wrote code before have begun building their own apps. In the coming years, millions more will!
I believe the majority of new apps will be web apps. Native is great, but there's way too much friction for casual developers.
For this reason, I am ultra bullish on browsers. The web will be the distribution platform of a billion empowered casuals 🤓
"Procedural metaprogramming where you can use the full language has nothing over simple textual substitution."
What are we even doing here?
Manuel Simoni retweeted
The "rewrite it in Goose" movement is already gaining steam, see here some early results of rewriting expertly written ninja/json/yaml/jq parsing in Goose and getting large speedups: github.com/calpeyser/goose-e…
One of the weirdest effects of LLMs is that people act as if writing extremely complex programs is somehow solved now.
Recently, I was talking about writing a compiler on here, and one of you suggested to "simply" write a verified compiler. As if writing a verified compiler is something that doesn't require you to be a top scientist AND crack programmer at the same time.
Somehow LLMs will do that for you, apparently. Despite there being like 1 verified compiler (CompCert) in existence at the moment. But somehow it's something that one simply does now.
Manuel Simoni retweeted
Invent type system
Make it unsound so that it’s very hard to be sure how edge cases are handled
Keep hallucinating new implementations back and forth
Make wild claims about performance
Can’t actually tell me if it type checks the same things
Mfw
I don't understand posts like this.
If you are motivated by the art or craft, nothing has (or can) be taken from you.
In fact, you've just gained a new tool that, while pretty mid overall, certainly increases your power.
Not really clued into gaming, but I don't think this is a good comms strategy.
How should you _not_ interpret this and the following tweet as "We threw together something shitty. For some reason it's fast lol. Will you test it for us"?
We vibe coded the Epic Games Launcher to be faster than the other guy.
Don't believe us? Test it out for yourself in Preview:
launcher.epicgames.com
Manuel Simoni retweeted
A fragment of OpenAI’s Navier-Stokes proof in Lean and the same translated by Astra to Naproche-ZF, which uses ForTheL for surface syntax (yes, you are looking at a formal language, not natural prose) and a version of set theory for semantics.
There should be a ForTheL frontend for Lean to lessen the pain of reading slop and unite the English proof and the code proof in one single artifact. That way there will be no mismatch scandals. But if ForTheL becomes the standard, Lean might not be necessary after all; any good backend with a better understood metatheory will do.
> I don't understand the code the LLM wrote for my CRUD app
> I know, extremely high-end techniques like formal verification and dependent types will fix me
Is there corposlop? No shit.
But read some code written by people who love what they're doing. Like Emacs, or Photoshop 1.0, MacDraw, QuickDraw. It's art.
(Not anti-LLM, I think they are great for producing exactly that same kind of quality, but not as a slop cannon.)
people mourning the death of handwritten software is the ultimate status signaling trap.
have you actually interacted with human-written codebases out in the wild? 99% of legacy software is an unreadable tower of bugs, visual slop, and terrible UX held together by spit and duct tape. it’s pure junk optimized for corporate bloat cycles.
the romanticization of the manual programmer typing out every single character line-by-line is bizarre. we are replacing a fundamentally flawed human bottleneck with flawless, deterministically optimized intelligence loops. let’s go!
I don't understand the Unison language. "Code is content-addressed" is at best an implementation detail. It doesn't change anything fundamental. The real issues of naming etc are entirely unaffected by this.
