twoodfin

10 ms·

301.
by twoodfin 6mo ago
Why? HN isn’t “curating” the wave of AI-written tech article slop. Unclear if they should, readers here love it! Hard to believe these models won’t get better and better at producing music that humans want to listen to.
302.
by twoodfin 6mo ago
This is really just a question of product design meeting the technology. Today, lots of integer compute happens on local devices for some purposes, and in the cloud for others. Same is already true for matmul, lots of FLOPS being spent loca
303.
by twoodfin 6mo ago
Wow, that’s a factoid I’d love to learn more about!
304.
by twoodfin 6mo ago
I don’t need a tool to tell me that, and if it was a well-written, interesting, accurate essay, I wouldn’t care. But it’s none of those three things. It is, however, the result of a model trained very effectively to give humans—including hn
305.
by twoodfin 6mo ago
LLMs are weaponized Gell-Mann amnesia when it comes to writing for humans.
306.
by twoodfin 6mo ago
I’m not saying it wasn’t possible, I’m saying the larger ecosystem was never going to embrace a language that was as heavyweight as Ada. In 1983, most PC system software was written in assembly!
307.
by twoodfin 6mo ago
LLMs do not currently improve the style of typical HN writing. Maybe someday they will; this article is less painfully bad than those of a few months ago. The main problem with this article is that it appears to have been basically written
308.
by twoodfin 6mo ago
I’ve never directly played with Ada but my understanding is that it was very much both. Ada includes a number of critical abstractions that require either dynamic runtime code (slow runtime) or the proverbial sufficiently smart compiler (sl
309.
by twoodfin 6mo ago
Neither of those standards are what I’m talking about. Obviously this article was highly pleasing to the hn audience as it’s currently sitting at #1. It’s still garbage, because it doesn’t have any interesting ideas behind it. Certainly not
310.
by twoodfin 6mo ago
The evidence is that the article’s writing is terrible. It repeats the same rhetorical devices over and over, dressing up a series of facts in false profundity, because there’s no actual authorial insight here. It’s just “write a well-resea
311.
by twoodfin 6mo ago
Ada’s failure to escape its niche is overdetermined. Given the sophistication of the language and the compiler technology of the day, there was no way Ada was going to run well on 1980’s microcomputers. Intel built the i432 “mainframe on a
312.
by twoodfin 6mo ago
The debates over these proposed OS mandates remind me of the Clinton-era Clipper chip: https://en.wikipedia.org/wiki/Clipper_chip May they meet a similar fate!
313.
by twoodfin 6mo ago
In the early days of socialization on the Internet it had a very different meaning!!
314.
by twoodfin 6mo ago
That is a really tough business if you can't match AWS' efficiency & reliability at scale. Presumably AWS also wants to be the AI version of AWS. (Amazon + Anthropic does seem like a much more compelling enterprise collaborati
315.
by twoodfin 7mo ago
Corporate contributions to Federal politicians and candidates are illegal in the US. The New York Times is allowed to spend money like anyone else praising or slagging politicians, but that’s the First Amendment, not funding candidates.
316.
by twoodfin 7mo ago
Yes, my opinion is that the average highly upvoted deep dive technical article on HN was substantially better in 2023 than in 2026. I also believe that the bulk of that degrade is due to frontier LLM’s currently being poorly trained to writ
317.
by twoodfin 7mo ago
The canonical industrial explanation “why not” is probably this 2010 piece from Joe Duffy @ Microsoft: http://joeduffyblog.com/2010/01/03/a-brief-retrospective-on-...
318.
by twoodfin 7mo ago
Your results match similar analyses of database systems I’ve seen. 64KB-128KB seems like the sweet spot.
319.
by twoodfin 7mo ago
I’m sorry, which specific people were “ground into paste” and when?
320.
by twoodfin 7mo ago
No, they don't, and I don't try to judge a book by its cover, anyway. TBC, I don't dislike LLM-written articles because I think the author is being lazy, and I can even live with the LLM-isms. I dislike LLM-written articles i
321.
by twoodfin 7mo ago
This is 100% AI slop. It’s really not obvious to you? Look at the rest of this blog: https://piechowski.io/post/ Almost no posts since 2020, then a swarm of LLM-style clickbait titles… Oh wait, the 2020 one also has a
322.
by twoodfin 7mo ago
I’m SOO happy but also wistfully sad when I open a post like this that I am desperately excited to read and it’s not muddled-thinking- and LinkedInese-riddled slop.
323.
by twoodfin 7mo ago
There’s a great video of a NeXT-era Steve Jobs keynote floating around—I think the one where he announces the x86 port as NeXT was transitioning to a software-only company—where he specifically calls out DriverKit and how great it is. Steve
324.
by twoodfin 7mo ago
Indeed, no taste.
325.
by twoodfin 7mo ago
No, I get it, and the benefits of putting what would have required a professional VFX team at (eventually) the fingertips of every amateur filmmaker are amazing. It’s also alarming what’s possible when reassembling photons en masse become
326.
by twoodfin 7mo ago
Sure but where the puck is going is long-running reasoning agents where local models are (for the moment) significantly constrained relative to a Claude Opus 4.6.
327.
by twoodfin 7mo ago
The idea of applying this modern magic to history & art is horrifying. The dream of Minitrue! Presumably Netflix wants to erase smoking from its back catalog or some other bit of papier-mâché Stalinism. Oh well, neat bit of auto-regress
328.
by twoodfin 7mo ago
That these data centers can turn electricity + a little bit of fairly simple software directly into consumer and business value is pretty much the whole story. Compare what you need to add to AWS EC2 to get the same result, above and beyond
329.
by twoodfin 7mo ago
I would expect that. But I’d also expect the pattern of their thoughts to look more varied in structure like C or German, and less like totally uniform s-expressions.
330.
by twoodfin 7mo ago
Sure, no argument that s-expressions are wonderfully simple & expressive. But most human languages—or at least the dominant ones that compose the vast bulk of the LLM training set—use more complex structuring rules for whatever evolutio

More ›