Andrej Karpathy talks about "Claws"

269 points by helloplanets 12 hours ago on hackernews | 64 comments

The actual content: https://xcancel.com/karpathy/status/2024987174077432126

Why is this linking to a blog post of what someone said, instead of directly linking to what they said?

[1] https://x.com/karpathy/status/2024987174077432126

Because the author of the blog is paid to post daily about nothing but AI and needs to link farm for clicks and engagement on a daily basis.

Most of the time, users (or the author himself) submit this blog as the source, when in fact it is just content that ultimately just links to the original source for the goal of engagement. Unfortunately, this actually breaks two guidelines: "promotional spam" and "original sourcing".

From [0]

"Please don't use HN primarily for promotion. It's ok to post your own stuff part of the time, but the primary use of the site should be for curiosity."

and

"Please submit the original source. If a post reports on something found on another site, submit the latter."

The moderators won't do anything because they are allowing it [1] only for this blog.

[0] https://news.ycombinator.com/newsguidelines.html

[1] https://news.ycombinator.com/item?id=46450908

geeunits | 11 hours ago

I've been warned for calling this out, but I'm glad others are privy to the obvious

PacificSpecific | 11 hours ago

Yeah it's really quite annoying. Is there a way to just block his site source from showing up on here without using external tools?

bahmboo | 11 hours ago

I find is very easy to hit the hide button. It makes reading the site much faster but there is some feeling of fomo.

: PacificSpecific | 11 hours ago
That's per-post though isn't? I can't ban a submission source can I?
Regardless thanks for the tip

CamperBob2 | 5 hours ago

What's wrong with external tools? Just ask Claude to vibe-code you a Simonblocker.

nl | 11 hours ago

Simon's work is always appreciated. He thinks through things well, and his writing is excellent.

Just because something is popular doesn't make it bad.

sunaookami | 11 hours ago

He massively fell off, is now only in for the marketing hype and even has a sponsor now for his blog. Sad.

UncleMeat | 10 hours ago

"Self promotion is allowed if your content is sufficiently good" is odd.

smallerize | 8 hours ago

Self-promotion is allowed. Doesn't even have to be good.

verdverm | 6 hours ago

The HN guidelines say don't use HN "primarily" for self promotion, which Simon does not do. He's an active member of the HN community.

owebmaster | 5 hours ago

He's an active member primarily self promoting

: verdverm | 4 hours ago
Simon's comment history indicates otherwise

owebmaster | 5 hours ago

That's not Simon's work or even any work at all, it is a link to a xit

hizanberg | 11 hours ago

So everyone has to waste their time to visit a link on a blog first instead of being able to go directly to the source?

and why would anyone down vote you for calling this out, like who wants to see more low effort traffic-grab posts like this?

: bahmboo | 11 hours ago
Because he didn't submit it.

bahmboo | 11 hours ago

The author didn't submit this to HN. I read his blog but I'm not on X so I do like when he covers things there. He's submitted 10 times in last 62 days.

bakugo | 11 hours ago

> He's submitted 10 times in last 62 days.

Now check how many times he links to his blog in comments.

Actually, here, I'll do it for you: He has made 13209 comments in total, and 1422 of those contain a link to his blog[0]. An objectively ridiculous number, and anyone else would've likely been banned or at least told off for self-promotion long before reaching that number.

[0] https://hn.algolia.com/?dateRange=all&page=0&prefix=true&que...

bahmboo | 10 hours ago

I like being able to follow tangents and related topics outside the main comment thread so generally I appreciate when people do that via a link along with some context.

But this isn't my site and I don't get to pick the rules.

npilk | 5 hours ago

So about 1 in 10? Doesn’t seem that terrible to me. Especially when many of them are in response to questions about his work, and he’s answering with a link to a different post.

I think 7 or 8 out of 10 would be a bad look.

: owebmaster | 4 hours ago
It depends.
How many of the comments without links were in a thread that started from the links? I'd guess at least some 2 or 3 out of 10.
What about just last year? We are probably close to 7 out of 10.
It's annoying.

Barbing | 5 hours ago

>anyone else

Perhaps not other thought leaders.

I would be curious to know:

How many clicks out from HN, and much time on page on average (on his site), and much subsequent pro-social discussion on HN, did those links generate versus the average linkout here? Wouldn’t change the rules but I do suspect[0] it would repaint self-promotion as something more genuine.

: bakugo | 4 hours ago
> Perhaps not other thought leaders.
Nice way of saying grifters.
You're bringing up essentially the same non-argument that dang himself used when he recently personally told off someone else for pointing out the same rule breaking behavior. It boils down to "People upvote it and comment on it so it must be good content regardless of which rules it breaks" which is a harmful way of thinking, the social media version of "laws are only for the poor".
If getting enough upvotes and replies elevates one above the rules, it should be clearly stated in said rules, but I have a feeling it never will be because it's obviously not a good look.

greenie_beans | 5 hours ago

he adds an insane amount of signal. some folks just can't look at the light and that's ok!

odshoifsdhfs | 11 hours ago

Hah i didn’t see who submitted it but as soon as I read your message i thought it was simonw, and behold, tada!

HN really needs a way to block or hide posts from some users.

consumer451 | 10 hours ago

Ironically, you could probably generate a browser extension or user script to do that in one to three prompts.

: agmater | 10 hours ago
If you can't one-shot that you've been declawed /s

duskdozer | 9 hours ago

firefox usercss or stylus addon, enjoy ;), no LLM needed

    tr.submission:has(a[href="from?site=<...>"])
    {
        display: none;

        & + tr
        {
            display: none;
        }
    }

    .comtr:has(.hnuser[href="user?id=<...>"])
    {
        display: none;
    }

This isn't just a CSS snippet—it's a monumentous paradigm shift in your HN browsing landscape. A link on the front page? That's not noise anymore—that's pure signal.

time to take a shower after writing that

manarth | 9 hours ago

HN formatting isn't quite markdown: you want a 4-space prefix to identify/format text as code.

duskdozer | 9 hours ago

my tabs :(

does it look measurably different this way? to me it looks the same but now indented

: manarth | 9 hours ago
Looks great now!
And thanks for an example with nested CSS, I hadn't seen that outside SASS before, hadn't realised that had made its way into W3C standards :-)

manarth | 9 hours ago

I described an approach here – feel free to use this if it's fit for your use-case:

https://news.ycombinator.com/item?id=46341604

simonw | 7 hours ago

But I didn't submit this.

odshoifsdhfs | 7 hours ago

It wasn't about the submission itself, is just about every post/comment you do about AI. I don't downvote you or anything, but a bit tired. So if it can save me time to just skip over submissions/comments I will do.

(for the rest, I was able to hide in Safari using manarth comment here: https://news.ycombinator.com/item?id=46341604

If anyone has one that will also work for user comments I would appreciate it.

simonw | 6 hours ago

Also write about rare New Zealand parrots and their excellent breeding season. Those posts don't tend to make HN though! https://simonwillison.net/tags/kakapo/

: greenie_beans | 5 hours ago
i very much appreciate your reporting on AI, please don't stop

Zetaphor | 5 hours ago

For what it's worth I enjoy your writing and commentary.

dandrew5 | 5 hours ago

I use a bookmarklet for this https://dan-lovelace.github.io/hn-blocklist/. Just added simonw's website to the blocklist as well.

Der_Einzige | 11 hours ago

Thank you for calling this out. The individual in question is massively overhyped.

[OP] helloplanets | 9 hours ago

> Because the author of the blog is paid to post daily about nothing but AI and needs to link farm for clicks and engagement on a daily basis.

Care to elaborate? Paid by whom?

throwup238 | 9 hours ago

It’s at the top of the page:

> Sponsored by: Teleport — Secure, Govern, and Operate AI at Engineering Scale. Learn more

https://simonwillison.net/2026/Feb/19/sponsorship/

: [OP] helloplanets | 9 hours ago
Ah, thanks. Somehow missed that.

simonw | 7 hours ago

> Most of the time, users (or the author himself) submit this blog as the source, when in fact it is just content that ultimately just links to the original source for the goal of engagement.

I encourage you to look at submissions from my domain before you accuse me like this: https://news.ycombinator.com/from?site=simonwillison.net - the ones I submitted list "simonw" as the author.

I'm selective about what I submit to Hacker News. I usually only submit my long-form pieces.

In addition to long form writing I operate a link blog, which this Claw piece came from. I have no control over which of my link blog pieces are submitted by other people.

I still try to add value in each of my link posts, which I expect is why they get submitted so often: https://simonwillison.net/2024/Dec/22/link-blog/ - in this case the value add was highlighting that this is Andrej helping coin yet another new term, something he's very good at.

yunohn | 5 hours ago

> Andrej helping coin yet another new term, something he's very good at

Ignoring all the other stuff, isn't this just a phenomenon of Andrej being worshipped by the AI hype crowd? This entire space is becoming a deification spree, and AGI will be the final boss I guess.

Barbing | 5 hours ago

Coining terms affects normies too, it hits all of our headlines and lexicons.

: yunohn | 5 hours ago
Completely agreed - and that media exposure is a result of clickbait journos piggybacking on the AI hype crowd. It's all a quite disappointing feedback loop.

simonw | 5 hours ago

Language matters. If you have a term that's widely understood you can have much more productive conversations about that concept.

"Agent" is a bad term because it's so vaguely defined that you can have a conversation with someone about agents and later realize you were both talking about entirely different things.

I'm hoping "Claw" does better on that basis because it ties to a more firm existing example and it's also not something people can "guess" the meaning of.

: yunohn | 5 hours ago
What is the firm example that provides meaning to “claw”? I guess we don’t have any concrete analytics, but I would be willing to bet that the fraction of people who actually used openclaw is abysmally small, vs the hype. “Agent”s have been used by a disproportionately larger number of people. “Assistant” is also a great existing term (understood by everyone), that encompasses what the blogs hyping openclaw discussed using it for as well.

Barbing | 5 hours ago

Honestly in the end, I hope you don’t change your behavior b/c you’re one of the most engaging and accessible writers in the loudest space on earth right now.

It is self-evident the spirit of no rule would intend to prohibit anything I’ve ever seen you do (across dozens and dozens of comments).

JKCalhoun | 8 hours ago

(Prefer the xcancel link [1] someone posted in this thread.)

[1] https://xcancel.com/karpathy/status/2024987174077432126

dcreater | 9 hours ago

[flagged]

thedevilslawyer | 9 hours ago

Rubbish. Simon is a good independent voice in capturing the llm zeitgeist.

blibble | 6 hours ago

Simon Willison claims to be an "Independent AI researcher"[1]:

but then at the top of this article:

> Sponsored by: Teleport — Secure, Govern, and Operate AI at Engineering Scale. Learn more

not exactly a coherent narrative, is it?

[1]: https://bsky.app/profile/simonwillison.net

simonw | 6 hours ago

I wrote a little note about that here - it even opens with "I value my credibility as an independent voice" https://simonwillison.net/2026/Feb/19/sponsorship/

I get (incorrectly) accused of writing undisclosed sponsored content pretty often, so I'm actually hoping that the visible sponsor banner will help people resist that temptation because they can see that the sponsorship is visible, not hidden.

blibble | 6 hours ago

> I value my credibility as an independent voice

not enough to not take their money though?

insipid

simonw | 6 hours ago

I'm currently planning to avoid sponsorship from companies that I regularly write about for that reason.

blibble | 6 hours ago

> I'm currently planning to avoid sponsorship from companies that I regularly write about for that reason.

ah so if it's not "regular" (which is completely arbitrary), then it's fine to call yourself independent while directly taking money from people you're talking about?

glad we cleared up the ambiguity around your ethical framework

: simonw | 6 hours ago
You're welcome to stop reading me if you think my ethics are irreversibly corrupted and you can no longer trust my writing.
Thankfully most of my readers are better at evaluating their information sources than you are.

akssassin907 | 3 hours ago

That's actually a cleaner editorial standard than most publications follow. The major risk in tech journalism isn't disclosed sponsorships — it's the undisclosed access journalism where coverage tone shifts to maintain relationships. Visible banners beat invisible influence every time.

simonw | 7 hours ago

You know I helped popularize "slop"? I get credited by Wikipedia as an "early champion": https://en.wikipedia.org/wiki/AI_slop

dang | 3 hours ago

Would you please not cross into personal attacks on HN? It's not what this site is for, and destroys what it is for. We've already had to ask you this, and we end up banning accounts that keep breaking the site guidelines this way.

https://news.ycombinator.com/item?id=46715512 (Jan 2026)

https://news.ycombinator.com/item?id=45022369 (Aug 2025)

If you wouldn't mind reviewing https://news.ycombinator.com/newsguidelines.html and taking the intended spirit of the site more to heart, we'd be grateful.

dang | 4 hours ago

Comments* moved to https://news.ycombinator.com/item?id=47096253, which has the original source.

We will add the current link to the toptext there as well.

(* except for the ones that only make sense in current context - that's the intention at least)