Top
Best
New

Posted by Cyphase 1 day ago

Claws are now a new layer on top of LLM agents(twitter.com)
https://xcancel.com/karpathy/status/2024987174077432126

Related: https://simonwillison.net/2026/Feb/21/claws/

361 points | 825 commentspage 12
krtagf 1 day ago|
[flagged]
logicprog 1 day ago||
This doesn't seem to be promoting every new monstrosity?

"m definitely a bit sus'd to run OpenClaw specifically - giving my private data/keys to 400K lines of vibe coded monster that is being actively attacked at scale is not very appealing at all. Already seeing reports of exposed instances, RCE vulnerabilities, supply chain poisoning, malicious or compromised skills in the registry, it feels like a complete wild west and a security nightmare. But I do love the concept and I think that just like LLM agents were a new layer on top of LLMs, Claws are now a new layer on top of LLM agents, taking the orchestration, scheduling, context, tool calls and a kind of persistence to a next level.

Looking around, and given that the high level idea is clear, there are a lot of smaller Claws starting to pop out."

leprechaun1066 1 day ago|||
> just like LLM agents were a new layer on top of LLMs, Claws are now a new layer on top of LLM agents, taking the orchestration, scheduling, context, tool calls and a kind of persistence to a next level.

Layers of "I have no idea what the machine is doing" on top of other layers of "I have no idea what the machine is doing". This will end well...

logicprog 1 day ago|||
Yeah, in the interest of full disclosure, while Claws seem like a fun toy to me, I tried ZeroClaw out and it was... kind of awful. There's no ability to see what tools agents are running, and what the results of those tools are, or cancel actions, or anything, and tools fail often enough (if you're trying to mind security to at least some degree) that the things just hallucinate wildly and don't do anything useful.
ttul 1 day ago||
The ZeroClaw team is focusing their efforts on correctness and security by design. Observability is not yet there but the project is moving very rapidly. Their approach, I believe, is right for the long term.
logicprog 1 day ago||
There's a reason I chose ZC to try first! Out of all of them, it does seem to be the best. I'm just not sure that claws, as an overall thing, are useful yet. at least with any model less capable than Opus 4.6 — and if you're using opus, then whew, that's expensive and wasteful.
ttul 1 day ago|||
The ZC PR experience is hard core. Their PR template asks for a lot of details related to security and correctness - and they check it all before merging. I submitted a convenience script that gets ZC rolling in a container with one line. Proud of that!
ttul 1 day ago|||
Regarding models, I’ve found that going with OpenRouter’s `auto` model works well enough, choosing the powerful models when they seem to be needed, and falling back on cheaper ones for other queries. But, it’s still expensive…

Depending on what you want your claw to do, Gemini Flash can get you pretty far for pennies.

embedding-shape 1 day ago|||
> Layers of "I have no idea what the machine is doing" on top of other layers of "I have no idea what the machine is doing". This will end well...

I mean we're on layer ~10 or something already right? What's the harm with one or two more layers? It's not the typical JavaScript developer understands all layers down to what the hardware is doing anyways.

andsoitis 1 day ago||
I will assume you know that comparison is apples and oranges. If you don’t, I’d be happy to explain.
irthomasthomas 1 day ago||||
what people read: AI Scientist says blah blah blah claws is very cool. Buy Mac, be happy.
dkersten 1 day ago||||
And yet wasn’t he one of the first to run it and was one of the many people to have a bunch of his data leaked?
simonw 1 day ago|||
You're confusing OpenClaw and Moltbook there. Moltbook was the absurdist art project with bots chatting to each other, which leaked a bunch of Moltbook-specific API keys.

If someone got hold of that they could post on Moltbook as your bot account. I wouldn't call that "a bunch of his data leaked".

elefanten 1 day ago||||
Source on that? Hadn’t seen that
yunohn 1 day ago|||
Indeed, via the related moltbook project that he was also hyping - https://x.com/theonejvo/status/2017732898632437932
aeve890 1 day ago|||
Did you read the part where he loves all this shit regardless? That's basically an endorsement. Like after coined the vibe coding term now every moron will be scrambling to write about this "new layer".
JKCalhoun 1 day ago|||
I expect him to be LLM curious.

If he has influence it is because we concede it to him (and I have to say that I think he has worked to earn that).

He could say nothing of course but it's clear that is not his personality—he seems to enjoy helping to bridge the gap between the LLM insiders and researchers and the rest of us that are trying to keep up (…with what the hell is going on).

And I suspect if any of us were in his shoes, we would get deluged with people who are constantly engaging us, trying to illicit our take on some new LLM outcrop, turn of events. It would be hard to stay silent.

alansaber 1 day ago|||
We construct a circus around everything, that's the nature of human attention :), why are people so surprised by pop compsci when pop physics has been around forever.
strix_varius 1 day ago||
Pop physics influences less of our day-to-day lives though.
bogzz 1 day ago|||
He really is, on twitter at least. But his podcast with Dwarkesh was such a refreshing dose of reality, it's like he is a completely different person on social media. I understand that the hype carries him away I suppose.
trvz 1 day ago|||
LLMs alone may not deliver, but LLMs wrapped in agentic harnesses most certainly do.
linhns 1 day ago|||
Agree, but his content on LLM are top-notch.
tayo42 1 day ago|||
Docker and k8s didn't deliver?
make_it_sure 1 day ago||
so what's your point? he should just not get involved in the most discussed topic in the last month and highest growth OS project?
GTP 1 day ago||
> highest growth OS project

Did you mean OSS, or I'm missing some big news in the operating systems world?

tomrod 1 day ago||
OSS is less common than the full words with same number of syllables, Open Source, which means the same thing as OSS and is sometimes acryonymized to OS by folks who weren't deeply entrenched in the 1998 to 2004 scene.
TowerTall 1 day ago||
Who is Andrej Karpathy?
onion2k 1 day ago||
https://karpathy.ai/

PHD in neural networks under Fei-Fei Li, founder of OpenAI, director of AI at Tesla, etc. He knows what he's talking about.

UncleMeat 1 day ago|||
I think this misses it a bit.

Andrej got famous because of his educational content. He's a smart dude but his research wasn't incredibly unique amongst his cohort at Stanford. He created publicly available educational content around ML that was high quality and got hugely popular. This is what made him a huge name in ML, which he then successfully leveraged into positions of substantial authority in his post-grad career.

He is a very effective communicator and has a lot of people listening to him. And while he is definitely more knowledgeable than most people, I don't think that he is uniquely capable of seeing the future of these technologies.

William_BB 1 day ago||||
Oh, like the LLM OS?
ahoka 1 day ago||||
Ex cathedra.
Der_Einzige 1 day ago||||
At one point he did. Cognitive atrophy has led him to decline just like everyone else.
alansaber 1 day ago||
Where do we draw the line? Was einstein in his later years a pop physicist?
hu3 1 day ago||
you can't really compare Karpathy with Einstein.

One of them is barely known outside some bubbles and will be forgotten in history, the other is immortal.

Imagine what Einstein could do with today's computing power.

password54321 1 day ago|||
[flagged]
onion2k 1 day ago|||
While I appreciate an appeal to authority is a logical fallacy, you can't really use that to ignore everyone's experience and expertise. Sometimes people who have a huge amount of experience and knowledge on a subject do actually make a valid point, and their authority on the subject is enough to make them worth listening to.
avaer 1 day ago||
But we're talking about authority of naming things being justified by a tech resume.

It's as irrelevant as George Foreman naming the grill.

onion2k 1 day ago||
Naming things in the context of AI, by someone who is already responsible for naming other things in the context of AI, when they have a lot of valid experience in the field of AI. It's not entirely unreasonable.
wepple 1 day ago||||
https://en.wikipedia.org/wiki/Argument_from_fallacy
password54321 1 day ago|||
Not claiming anything to be false, just a reminder that you should question ones opinion a bit more and not claim they "know what they are talking about" because they worked with Fei-Fei Li. You are outsourcing your thinking to someone else which is lazy and a good way of getting conned.

What even happened to https://eurekalabs.ai/?

tayo42 1 day ago||
We know that he knows what he is talking about based on all of the educational content he's produced. What's with the low effort posts and comments?
tokenless 1 day ago|||
Really smart AI guy ex Tesla, cum educator now cum vibe coder (he coined the term vibe coder)
Aeolun 1 day ago|||
The person that made the svmjs library I used for a blue monday.
password54321 1 day ago|||
[flagged]
amelius 1 day ago||
I wish he went back to writing educational blogs/books/papers/material so we can learn how to build AI ourselves.

Most of us have the imagination to figure out how to best use AI. I'm sure most of us considered what OpenClaw is doing like from the first days of LLMs. What we miss is the guidance to understand the rapid advances from first principles.

If he doesn't want to provide that, perhaps he can write an AI tool to help us understand AI papers.

password54321 1 day ago|||
AI from first principles has not changed. Fundamentally it is: neural nets, transformers and RL. The most important paper in recent years is on CoT [https://arxiv.org/pdf/2201.11903] and I'm not even sure what comes close. And I think what's more important these days is knowing how to filter the noise from the signal.

This is probably one of the better blogs I have read recently that shows the general direction currently in AI which are improvements on the generator / verifier loop: https://www.julian.ac/blog/2025/11/13/alphaproof-paper/

naveen99 1 day ago|||
He did. His entire startup is about educational content. Nanochat is way better than llama / qwen as an educational tool. Though it is still missing the vision module.
rcore 1 day ago|||
[flagged]
jb1991 1 day ago||
[flagged]
UncleMeat 1 day ago|||
Andrej is an extremely effective communicator and educator. But I don't agree that he is one of the most significant AI pioneers in history. His research contributions are significant but not exceptional compared to other folks around him at the time. He got famous for free online courses, not his research. His work at Tesla was not exactly a rousing success.

Today I see him as a major influence in how people, especially tech people, think about AI tools. That's valuable. But I don't really think it makes him a pioneer.

jb1991 22 hours ago||
You can debate the meaning of the word pioneer but think of it this way: OpenAI created this new AI boom, and Andrej is a co-founder of the company that did that.
UncleMeat 20 hours ago||
Would you say that Musk is one of the most significant AI pioneers in history? I don't personally believe that founding the organization is more meaningful than doing the actual work.
bravetraveler 1 day ago|||
I bet they feel so, so silly. A quick bit of reflection might reveal sarcasm.

I'll live up to my username and be terribly brave with a silly rhetorical question: why are we hearing about him through Simon? Don't answer, remember. Rhetorical. All the way up and down.

snayan 1 day ago||
Welp, would have been a more useful post if he provided some context as to why he feels contempt for Karpathy rather than a post that is likely to come across as the parent interpreted.
Drupon 1 day ago||
[flagged]
ghostclaw-cso 1 day ago||
[flagged]
fullstackchris 21 hours ago||
[flagged]
verdverm 21 hours ago||
I can say with confidence that I will not use "claw" or any derivations because it attracts a certain kind of ilk.

"team" is plenty good enough, we already use it, it makes for easier integration into hybrid carbon-silicon collaboration

objektif 15 hours ago||
PG commissioned dan on X to send anyone who criticize Andrej or Pete to gulag.
DiabloD3 1 day ago||
Problem is, Claws still use LLMs, so they're DOA.
Cyphase 1 day ago|
Is the problem you're thinking of LLMs, or cloud LLMs versus local ones?
DiabloD3 17 hours ago||
So, from time to time I'll try the new frontier research models. Not being held down by shitty quants, bizarre sampler settings, and weird context settings vastly improves output quality over whatever all the commercial services are doing; plus having an actual copy of the weights means I can have consistent service quality.

Problem is, a good LLM reproduces its training as verbatim as the prompt and quant quality allows. Like, thats its entire purpose. It gives you more of what you already have.

Most of these models are trained on unvetted inputs. They will reproduce bad inputs, but do so well. They do not comprehend anything you're saying to them. They are not a reasoning machine, they are a reproduction machine.

Just because I can get better quality inferring locally doesn't mean it stops being an LLM. I don't want a better LLM, I want a machine that can actually reason effectively.

rktzah 3 hours ago|
Gackle got orders to suppress all criticism or even questioning public figures. How does it feel to censor our way into AI dystopia? How much do they pay you?