Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Nvidia NemoClaw

github.com

Post reply on HN
Print view View original

231–240 of 291 posts

Re: Nvidia NemoClaw

#231

Post by dustypotato »

jychang wrote: post #119
Someone1234 wrote: post #76

Earlier quoted context omitted.

I think the point you're making is fully correct, so consider this a devil's advocate argument... People claim, you can use Claw-agents more safely while getting some of the benefits, by essentially proxying your services. For example on Gmail people are creating a new Google accounts, forwarding email via rule, and adding access to their calendar via Google's Family Sharing. This allows the Claw agent to read email,…

We already have this concept. It’s called user accounts. Your Gmail account vs my Gmail account. Your macOS account vs my macOS account. Yes, I can spam you from my Gmail. Yes, I can use sudo on my Mac and damage your account. But the impact is by default limited. The answer is to just treat assistants as a different user profile, use the same sharing mechanisms already developed (calendar sharing, etc), and call it…

Isn't this what the parent is saying ?

Re: Nvidia NemoClaw

#232

Post by vidarh »

Netcob wrote: post #65

Am I missing something? Why is everyone talking about sandboxes when it comes to OpenClaw? To me it's like giving your dog a stack of important documents, then being worried he might eat them, so you put the dog in a crate, together with the documents. I thought the whole problem with that idea was that in order for the agent to be useful, you have to connect it to your calendar, your e-mail provider and other servic…

Limiting the blast radius when a bomb goes off is still helpful even if you don't prevent the bomb from going off.

Now, you're right that sandboxing them is insufficient, and a lot of additional safeguards and thinking around it is necessary (and some of the risk can never be fully mitigated - whenever you grant authority to someone or something to act on your behalf, you inherently create risk and need to consider if you trust them).

Re: Nvidia NemoClaw

#233

Post by ahd94 »

yet another marketing stunt ? I like the framework that is used to buidl openclaw than open claw. Especially for work use cases, i would feel safer with a custom built version.

Re: Nvidia NemoClaw

#234

Post by ethbr1 »

Netcob wrote: post #228
observationist wrote: post #133

Earlier quoted context omitted.

What makes it even better is that these dogs are like Malinois. If they want to get into something, they will; people have had their entire network compromised by bots they left running overnight, and any important information like account logins and so on runs the risk of being misused. It's one thing to sandbox, maybe give the bot a temporary, limited $100 card or account to go perform a specific task, but there's…

I think it's a use case that identity/authorization/permission models are simply not made for. Sure, we can ban users and we can revoke tokens, but those assume that: 1. Something potentially malicious got access to our credentials 2. Banning that malicious entity will solve our problem 3. Once we did that, repaired the damage and improved our security, we don't expect the same thing to happen again None of these app…

> What we need is less like a sturdy door with a fancy lock, and more like that special spoon for people with parkinson's. Unlimited undo history.

Agree -- you can't solve probabilistic incorrectness with redresses designed for deterministic incorrectness.

This is like the 'How i parse html w regex?' question.

Imho, the next step is going to be around human-time-efficient risk bounding.

In the same way that the first major step was correctness-bounding (automated continuous acceptance testing to make a less-than-perfect LLM usable).

If I had to bet, we'll eventually land on out-of-band (so sufficiently detached to be undetectable by primary LLM) stream of thought monitoring by a guardrail/alignment AI system with kill+restart authority.

Re: Nvidia NemoClaw

#235

Post by ethbr1 »

macNchz wrote: post #189
TheDong wrote: post #176

Earlier quoted context omitted.

> people have had their entire network compromised by bots they left running overnight I'm curious if you have references to this happening with OpenClaw using one of the modern Opus/Sonnet 4.6 models. Those models are a bit harder to fool, so I'm curious for specific examples of this happening so I can do a red-team on my claw. I've already tried all sorts of prompt injections against my claw (emails, github issues,…

No maliciousness or injection required, even the newest and most resistant models can start doing weird stuff on their own, particularly when they encounter something failing that they want to work. Just today I had Opus 4.6 in Claude Code run into a login screen while building and testing a web app via Playwright MCP. When the login popped up (in a self-contained Chromium instance) I tried to just log in myself with…

> it turned out Claude had run code to query superadmin users in the database, picked the first one, and changed the password to `password123` so it could log in on its own.

Man, every LLM quirk behavior really is a thing a monomaniacal junior dev would do...

Re: Nvidia NemoClaw

#236

Post by rune-dev »

mx_03 wrote: post #194
rune-dev wrote: post #138

Earlier quoted context omitted.

But isn’t this just another notification to ignore? The ticket being assigned to you is your “Hey take care of this!” ping, same with the email or text from your friend. How long until you start tuning out the openclaw notifications?

cant that be fixed, tho? If the analogy is a personal asistant, a good assistant will know when to notify you and when not to.

Maybe yeah, though I don’t know if practicing restraint is something I would say LLM’s are good at though.

I think to all of the needless comments in code, AI code reviews pointing out inane nitpicks, etc.

It just makes me think your AI assistant is going to be pinging you non stop

Re: Nvidia NemoClaw

#237

Post by pmdr »

sailfast wrote: post #121
CrzyLngPwd wrote: post #111

Gotta say, that I feel kind of sad for the people that feel the need for these claw things. Are they so busy with their lives that they need an assistant, or do they waste their lives speaking to it like it is a human, and then doomscrolling on some addictive site instead of attending to their lives in the real world?

It’s not a need - it’s a fun new thing - fun to see what’s possible and how it helps. OpenClaw is not easy to set up or user friendly for most (BlueBubbles and Claw had an annoying bug recently) - but the way I have seen it work well requires an up front time investment and then interest compounds RAPIDLY to help manage things and be more productive. My guess is maybe you’ve never had an assistant or tried a Claw ins…

LLMs have earned their place in many jobs, but I struggle to see claws as more than a rather expensive waste of time and tokens. Downsides are gargantuan, effects of dead internet theory will be ubiquitous.

Re: Nvidia NemoClaw

#238

Post by ethbr1 »

jychang wrote: post #119
Someone1234 wrote: post #76

Earlier quoted context omitted.

I think the point you're making is fully correct, so consider this a devil's advocate argument... People claim, you can use Claw-agents more safely while getting some of the benefits, by essentially proxying your services. For example on Gmail people are creating a new Google accounts, forwarding email via rule, and adding access to their calendar via Google's Family Sharing. This allows the Claw agent to read email,…

We already have this concept. It’s called user accounts. Your Gmail account vs my Gmail account. Your macOS account vs my macOS account. Yes, I can spam you from my Gmail. Yes, I can use sudo on my Mac and damage your account. But the impact is by default limited. The answer is to just treat assistants as a different user profile, use the same sharing mechanisms already developed (calendar sharing, etc), and call it…

That's punting the problem in the same way SELinux did. Agent loops are useful precisely because they're zero config.

Problem: I want to accomplish work securely.

Solution: Put granular permission controls at every interface.

New problem: Defining each rule at all those boundaries.

There's a reason zero trust style approaches won out in general purpose systems: it turns out defining a perfect set of secure permissions for an undefined future task is impossible to do efficiently.

Re: Nvidia NemoClaw

#240

Post by bigfishrunning »

BeetleB wrote: post #202
Kye wrote: post #193

Earlier quoted context omitted.

Okay. You have sensible escape prevention. Now this tool spreads. You help everyone get it set up. Someone hacks the site, injects a prompt lying about some event, maybe Drag Queen Story Hour in a place with lots of people enraged about it. Now there's chaos and confusion. Corrections chase the spread of misinformation.

> Now this tool spreads. You help everyone get it set up. Someone hacks the site You sound like my dad in the 90's, when it came to modems. Same tool. Good uses. Bad uses. The bad doesn't negate the good (c.f. Bittorrent).

I could make that same argument about giving my 9 year old a chainsaw and telling her to cut some wood

In the best case, some wood gets cut. There are many many worse things that can happen

But hey, same tool. Good uses. Bad uses.

Post reply on HN