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

Claws are now a new layer on top of LLM agents

twitter.com

Post reply on HN
Print view View original

991–1000 of 1001 posts

Re: Claws are now a new layer on top of LLM agents

#991

Post by rushcar »

lwhi wrote: post #982
ErneX wrote: post #953

Earlier quoted context omitted.

I know, but which open model that fits in there is useful enough for OpenClaw? I don’t think there is one. If you look at the videos and blog posts where they recommend getting a Mac Mini for this are recommending the base model (which comes with just 16GB), precisely because it’s the cheapest Mac that can read your reminders, use iMessage etc. that’s what those using OpenClaw want from the Mini, not its inference ca…

I disagree.

What model are you running with 64GB of VRAM (equivalent)? I doubt most users are doing that. Looking at their documentation, the default path for openclaw seems to be a 3P API for the model.

Re: Claws are now a new layer on top of LLM agents

#992

Post by hamburglar »

yencabulator wrote: post #990
hamburglar wrote: post #989

Earlier quoted context omitted.

If the “clawness” means you only use the llm to control itself, then yes, that’s impossible. But you can easily shim such a process so that the interfaces it uses to “claw out” to the real world are shims that have safeties such as human control. Openclaw does not do this, and is thus a scary shit show, but you can play with it in isolation safely, and I think a standard pattern for good control will emerge.

> easily Yeah that's an active research topic for teams of PhDs, including some of Google's brightest. And the current approach even with added barriers may just be fundamentally untrustable. Read the links from my earlier comment for background.

If the shim doesn’t use an LLM to make its decisions this is not a problem.

If the shim does use an LLM but no uncontrolled data is allowed in, this is not a problem.

Re: Claws are now a new layer on top of LLM agents

#993

Post by yencabulator »

hamburglar wrote: post #992
yencabulator wrote: post #990

Earlier quoted context omitted.

> easily Yeah that's an active research topic for teams of PhDs, including some of Google's brightest. And the current approach even with added barriers may just be fundamentally untrustable. Read the links from my earlier comment for background.

If the shim doesn’t use an LLM to make its decisions this is not a problem. If the shim does use an LLM but no uncontrolled data is allowed in, this is not a problem.

I think you're misunderstanding the severity of the lethal trifecta. Just because you put access controls around the LLM doesn't mean all that much if the access controls allow anything in & out. There is no way to write a shim that blocks "everything naughty", while remaining useful.

You literally have to fully prevent all outside input, or you have to prevent all exfiltration routes including web page reading (even the choice of links to follow is an exfiltration mechanism). At that point, what's left? What do you think will be on your allowlist?

I seriously doubt the early adopters of these software bundles use their assistants like with such restraint (https://xcancel.com/summeryue0/status/2025774069124399363), and that idealized image of these access control shims is not realistic.

Re: Claws are now a new layer on top of LLM agents

#994

Post by hamburglar »

yencabulator wrote: post #993
hamburglar wrote: post #992

Earlier quoted context omitted.

If the shim doesn’t use an LLM to make its decisions this is not a problem. If the shim does use an LLM but no uncontrolled data is allowed in, this is not a problem.

I think you're misunderstanding the severity of the lethal trifecta. Just because you put access controls around the LLM doesn't mean all that much if the access controls allow anything in & out. There is no way to write a shim that blocks "everything naughty", while remaining useful. You literally have to fully prevent all outside input, or you have to prevent all exfiltration routes including web page reading (even…

Your definition of “remaining useful” seems to require a lot more than mine. An email shim, for example could have destination whitelists, rate limits, an overall message quota, and can have its contents driven by fixed templates which the LLM can choose from, but not inject arbitrary data into. The point is that your claw need not have “do anything” powers, it needs to have extremely constrained powers. Maybe that is, as you say, “not a claw.” In fact, mine calls itself a “clav” because it’s almost a claw, but not quite.

Re: Claws are now a new layer on top of LLM agents

#995

Post by lwhi »

rushcar wrote: post #991
lwhi wrote: post #982

Earlier quoted context omitted.

I disagree.

What model are you running with 64GB of VRAM (equivalent)? I doubt most users are doing that. Looking at their documentation, the default path for openclaw seems to be a 3P API for the model.

It doesn't matter what 'most users' are doing.

On a 64 GB Apple silicon Mac mini you can natively host mid sized and some larger quantised local models .. using Ollama.

For example:

Qwen3-Coder (32B), GLM-4.7 (or GLM-4 Variants), Devstral-24B / Mistral Large (Quantized)

Re: Claws are now a new layer on top of LLM agents

#996

Post by ZitchDog »

aix1 wrote: post #744
ZitchDog wrote: post #484

Earlier quoted context omitted.

I've created my own "claw" running in fly.io with a pattern that seems to work well. I have MCP tools for actions that I want to ensure human-in-the loop - email sending, slack message sending, etc. I call these "activities". The only way for my claw to execute these commands is to create an activity which generates a link with the summary of the acitvity for me to approve.

Is there a risk that the summary doesn't fully match the action that actually gets executed?

Nope! The summary is presented to the user via a link and once the user follows the link and approves, the action is implemented entirely outside of the agent, on a separate server.

Re: Claws are now a new layer on top of LLM agents

#997

Post by ZitchDog »

devonkelley wrote: post #972
ZitchDog wrote: post #484

Earlier quoted context omitted.

I've created my own "claw" running in fly.io with a pattern that seems to work well. I have MCP tools for actions that I want to ensure human-in-the loop - email sending, slack message sending, etc. I call these "activities". The only way for my claw to execute these commands is to create an activity which generates a link with the summary of the acitvity for me to approve.

The approval-link pattern for gating dangerous actions is something I keep coming back to as well, way more robust than trying to teach the agent what's "safe" vs not. How do you handle the case where the agent needs the result of the gated action to continue its chain? Does it block and wait, or does it park the whole task? The suspend/resume problem is where most of these setups get messy in my experience.

I haven't needed that yet! But it seems like the agent could easily be notified of completed activities.

Re: Claws are now a new layer on top of LLM agents

#998

Post by michaelmrose »

Loughla wrote: post #905
michaelmrose wrote: post #872

Earlier quoted context omitted.

I really like the idea of multiple vectors although nobody seems apt to pick up the idea other than Slashdot.

The reason I left there was the down vote brigade that really killed most genuine criticism that disagrees with the sites pre formed opinions on certain topics. So I'm not sure it's a solved problem. Unless it's gotten better since 2011?

I think the specific community and some of the ways it does moderation and voting are seperable and I would love to see the latter tested on an open source discussion platform

Re: Claws are now a new layer on top of LLM agents

#999

Post by Gareth321 »

ai-christianson wrote: post #861
Gareth321 wrote: post #822

Earlier quoted context omitted.

It was surprisingly difficult for me to understand the use case as well. Here is my best attempt at an elevator pitch: At present your memories are proprietary data in whichever LLM you use. ChatGPT keeps all your conversations and output and data forever. What if you don't like GPT 5.2? What if you want to use other models as well? Or use the best model for the job? OpenClaw gives you that ability. Your memories and…

Gobii did always-on, connected to comms channels (sms, email), full computer/headed browser 8 months ago, in a much more secure and k8s-native way.

I agree it's pretty close, but it's proprietary, expensive, business focused, and more difficult to implement. It was never going to get the same consumer penetration of OpenClaw. Which is open sourced, almost no guardrails, easy to set up, and locally hosted.
Post reply on HN