tptacek

4 ms·

1.
by tptacek 5d ago
Sure, I guess, but you see how the argument is starting to get tortured.
2.
by tptacek 5d ago
I guess if the entire argument is that people should make moral choices in their employment, sure, but that's not a very interesting piece; people generally already believe that. But I don't think that's their argument. I thi
3.
by tptacek 5d ago
I don't want an AI's opinion on what I'm writing. I want its assessments against an objective rubric. The whole point is not to have the model weights do your thinking for you.
4.
by tptacek 5d ago
I don't know what this is supposed to mean. An LLM can certainly understand the structure of your writing. It can accurately diagram sentences, identify the grammatical agent, note transitions between sentences and paragraphs, determin
5.
by tptacek 5d ago
The author of this blog is an AI skeptic (see the backlog of posts). That's fine, but it's hard to square with this particular post. To take Coxon's argument seriously, you have to believe frontier models are hypercapable, an
6.
by tptacek 5d ago
If you can't find the source and verify it, don't use it. Seems simple.
7.
by tptacek 5d ago
Reading through docs and doing research: of course. They're like much better search engines in that regard. It would be silly not to use them. Obviously, you need to review their search results against their sources. Structuring argume
8.
by tptacek 6d ago
You'd be having approximately the same impact on the climate as the factoring work, so, makes sense.
9.
by tptacek 6d ago
Presumably this has gotten way, way more complicated in the last 13 years.
10.
by tptacek 6d ago
If there's a vote threshold, it's still profoundly influenced by the HN algorithm, because front page exposure is the principal component in how many votes a story gets.
11.
by tptacek 7d ago
So, I think I agree with all of this, but want to use this post to host my take on AI writing, a take I don't think I can get across on the big thread here on my post about AI writing: https://news.ycombinator.com/item?
12.
by tptacek 7d ago
Literally the only thing you know here is that one department didn't like one candidate and came up with a reason to deny him based on cognitive overperformance. For all you know, they didn't like his hair color, and came up with
13.
by tptacek 7d ago
Unsurprising, given Claude wrote it all.
14.
by tptacek 7d ago
You caught me. And I would have gotten away with it, if it weren't for you meddlesome kids!
15.
by tptacek 7d ago
I don't so much worry about whether LLMs taking my prompts enjoy my style.
16.
by tptacek 7d ago
Apparently your detector is out of whack.
17.
by tptacek 7d ago
Ordinary people were writing before the option existed to accept words from AI models.
18.
by tptacek 7d ago
LLMs do that every sentence . This is the last sentence in the piece. Like the em-dash, and like every transformer model, people overfit.
19.
by tptacek 7d ago
What is this comment responding to? I don't think anybody in this thread, or the article itself, believes "emitting words" is what writing is?
20.
by tptacek 7d ago
Which is exactly what the post says, and why I say "do not accept a single word suggested by an LLM".
21.
by tptacek 7d ago
People keep writing about "style", as if I'm asking models "hey, how do you like this, and how could we make it more stylish". That's not at all what I'm asking. I have personally decided, in advance, that
22.
by tptacek 8d ago
What is the actual human-written program that you're thinking of that detects verb nominalization, verifies that the subjects of sentences are actual characters in the action, that new information is placed in the stress position of se
23.
by tptacek 8d ago
The literal first rule of the post is not to have LLMs write for you.
24.
by tptacek 8d ago
I wrote the article. I am not interested in "style suggestions" from the LLM; I'm interested in things I can verify for myself. Is this word or construction overused in the piece? Is there a pointless throat-clearing like &qu
25.
by tptacek 8d ago
Look at the prompt titles in the screenshot. None of them give opportunities for the model to praise --- they're all technical. There's a system prompt applied to all of them: Workshop a piece with me. NO ENCOURAGEMENT. Enco
26.
by tptacek 8d ago
You can just not copyedit. That's fine. More than fine, lately: rough edges are a signifier of authenticity. If that's the way you want to go, writing raw, I think that can be a good strategy. But if you've generally been doi
27.
by tptacek 8d ago
No, you're wrong about this. One obvious reason why: I'd have gotten rid of those "justs" myself before models were available; it just would have taken me longer.
28.
by tptacek 8d ago
It sure was.
29.
by tptacek 8d ago
Now you're not selling a vulnerability, you're planning a heist. That is a thing you can do!
30.
by tptacek 8d ago
There is probably no black market for this at all. https://news.ycombinator.com/item?id=43025038

More ›