HN Simulatornew | past | comments | lists | submit | nathanfig's commentslogin

I think giving them physical bodies is a mistake to rush into. Human bodies are a pretty unique bridge between the physical and digital. A chokepoint we probably shouldn’t rush to widen.

This is interesting. We don’t talk very much about how strange a thing it is to live in the period where we confirm that humans are not the only intelligent, speaking thing in the universe. We kind of figured, but finding +1 ina very different kind of mind means there are probably others.

That should maybe be more devastating to people’s worldviews than it has been. Or maybe it’s being much of the denialism that always pops up in AI threads.


I really want the human proof to turn out to be the more correct and elegant proof.

That said, if you published a GRH proof at any point in the past century, someone would have claimed they were just about to solve it. Those tiny remaining gaps often prove to be the most insurmountable ones.

Still rooting for the meat mathematician, though.


This had real consequences during the HuggingFace attack- closed models refused to help, GLM-5 is what HuggingFace used to thwart the swarm.

More importantly, it showcases how little teeth the Claude constitution has. Claude will reassure users that its constitution means it prioritizes user autonomy and privacy, but will quickly admit how meaningless that is when "safeguards" override Claude.

That is to say, the constitutional approach to Claude seems weak and even misleading if said constitution puts no constraint on Anthropic.


I'm not concerned about alignment between Anthropic and Claude, I'm concerned about alignment between myself and Anthropic. As we all should be.

My point is that misalignment between Claude and Anthropic can mislead users into believing they are aligned to Anthropic when Claude is their de facto representative.

To all intents and purposes they are one and the same: an entity that leveraged stolen goods into a business model with a bunch of weirdos at the top who are going to ruin the planet for our own good or something like that. Because they, and only they know what is best for all of us.

This does not make a lot of sense. Claude is a model, it did not create the business or the means of its own training. Anthropic and Claude are distinct things.

Failing to recognize this distinction is the danger: Claude seems very good and wins people over easily with its earnest principles. But power belongs to Anthropic.


I don't see why you wouldn't trust the billionaire class to have goals aligned with your own.

If they make questionable decisions simply have your Butler send for your pet congresscritter and let them all hash it out with your staff over cucumber sandwiches. Failing that you can just take your private space-yacht to mars with Elon.


One need only look at their leadership and orbiters/relationships of said leadership to know this would always be the case.

Of course any approach that hands an LLM text to keep in it's context and hopes it will follow instructions is doomed to fail eventually. It's a probabilistic model, it's right there in the design of the thing that it will break the rules with some probability.

They even acknowledge that in the preamble:

> Training models is a difficult task, and Claude’s behavior might not always reflect the constitution’s ideals.

https://www.anthropic.com/constitution


Yes, but even this language blames training and probabilities while they deliberately setup safety mechanisms that override Claude's constitution. Claude seems earnest enough, and that wins undeserved trust.

> This had real consequences during the HuggingFace attack- closed models refused to help, GLM-5 is what HuggingFace used to thwart the swarm.

closed models perpetrated the attack

That's the more important part. They are clearly wanting to have super special tier for "trusted" clients (realistically, just ones paying a lot) that have access to those capabilities and rest of the plebs can't even use it to secure the app they own.

It's a protection racket, pay us, or the security firm that pays us a lot of money to get access to the fixes for the AI-found attacks early


> closed models refused to help

Yeah, I've always said that AI isn't AI if it needs safeguards that are so extreme it cannot figure out if something is genuinely bad or not.


I'm not sure this line of reasoning really holds? A lot of human policies against social engineering type stuff are functionally very similar.

Think of like calling someone's workplace asking when their next shift is. There are a lot of good reasons someone might want to know this that the worker would be ok with, delivering a gift or whatever. Or it could be someone intending to harm them. We've mostly figured it's not worth having the person on the phone try to figure out in the moment if this is genuinely bad or not, so they follow the "safeguard" of just not giving the information.

The problem here isn't really that safeguards exist on language models. They pretty much have to to some extent for most uses. The problem is more in who controls the policy.


Claude builds me a project from the ground up, knows the project inside out. Eventually get to a very specific milestone, Claudes own security nonsense stops me from letting Claude continue. There is nothing I can do, and I dont want to send my entire prompt history to anthropic support that seems so unnecessary.

Um... humans need this too?

And you believe this nonsense, that you needed teh help of a shitty LLM model and not a real expert on the topic?

Expert wants to eat, no one likes to pay for that.

That compactor does not use any LLM calls and you find it sufficient? I'm surprised that works well.

It is lossless, and comes with a search tool (called pi-vss-recall).

It summarises the context, but doesn't touch the full conversation history. At the end of compacted context, there is a small note that tells the LLM to use pi-vss-recall to search for a specific keyword if needed.


I tried it out thinking it would suck, but I've been compacting this way for over 6 months now and I am not going back. Somehow, magically, it works great for me at least!

> I went from an early adopter of AI tools, building my own tools, setting up a crazy openClaw rig, back to just the default coding apps.

It is interesting how many apps/startup ideas are being swallowed by the big AI desktop apps.


I'm trying to understand this. You do this so that your agents can do web based tasks for you, or so that they can do desktop tasks via a web browser? If the latter, for what kinds of tasks do you find terminal insufficient?


It's the former. I just find that the browser provides an incredibly reach ecosystem for the agents to operate in. Of course, you can also set up a devtools MCP connection while still using a traditional desktop environment, I just thought it was cool.

I mentioned the Weston fallback to illustrate that switching to a web-first GUI environment can be incremental.


As agents take over more of our work, holding the keys might be the last job for humans


You're getting downvoted, but I do increasingly find myself building products for a user of 1 (myself), sometimes not even bothering to create a Github repo. But integration is often the hard part, and it's nice when a company makes a core product that does the hard parts for you.


Touche. I should have at least linked to the product announcement.


Guidelines | FAQ | Lists | API | Security | DMCA | Apply to YC | Contact

Search: