Hacker News Reader

Comment by fasterik | Hacker News Reader

>I guess we probably want our tech to work exactly like this.

Do we, though? What if someone started an AI company that uses end-to-end encryption to make it impossible for anyone but you to access your data? Personally, I would switch to it in a heartbeat assuming it's competitive with the other products. I don't think it's the tech companies' job to surveil the population and prevent crimes. That said, I'm not necessarily against Anthropic or other companies reporting suspicious activity if their existing systems are detecting it. I'm just not sure we want every product to be forced into that data model.

Your follow-up about purely private tech seems to contradict your first statement. We can either have privacy or surveillance, not both.

View on HN

>What if someone started an AI company that uses end-to-end encryption to make it impossible for anyone but you to access your data? Personally, I would switch to it in a heartbeat assuming it's competitive with the other products.

Why wait for a company to build it? Get your own local hardware like I did and have those guarantees because YOU set it up.

I don't consider investing $20k into hardware to run SOTA open models, that are far behind proprietary SOTA, to be competitive.

Even if open models were competitive, it's still typically going to be more expensive than a cloud provider because of low utilization and higher purchase price.

How do I know that a private LLM system won't have a degradation of quality similar to this or worse? The only thing I can think of for the proposed scenario is some sort of homomorphic encryption system? But not sure.

>How do I know that a private LLM system won't have a degradation of quality similar to this or worse?

You use benchmarks, you test, and because YOU'RE the sysadmin you know what weights are running at what time, it's very visible. You can airgap the hardware and be guaranteed it won't change over time. And, frankly, degradation over time doesn't seem to be what's happening with the open models.

Last time I checked it costs closer to 500K to run SOTA open models at any usable speed.

Right now, I would be willing to pay ~$20 extra per month for strong privacy, assuming the same capabilities as Opus 5.5. I don't see local models making sense economically any time soon unless you value privacy at $1000's per month or are fine with much lower performance on hard tasks.

This already exists, confer.to, made by moxie (the guy that made Signal)

> Your follow-up about purely private tech seems to contradict your first statement.

That is correct! And exactly my point.

We want both, but, on paper, that is impossible. But in reality, we make it sorta mostly happen anyway, through making the easy defaults not private, and the private stuff not easy.

This is not ideal, because [various reasons I do not need to tell you], but it has proven to be the best we can do to mostly achieve both goals.

Kinda like how capitalism isn't great but just the least worst option we've found so far.

___

The actual fundamental underlying problem being that not all people are equal, but we kinda have to pretend they are, because not doing so leads to fascism and other terrible stuff.

But we kinda also do not want to fully pretend that, because doing so leads to yet other terrible stuff.

Hence the quadruple-speak and contradictions to kinda sorta somehow have a somewhat functioning reality.

That is kind of rambling. Our rambling score says we should observe you.

Oh don't worry, I'm certainly already on various lists, but the observations also will have resulted in the assessment that I am stable and no threat :)

I think we can have both and it might even be smart.

A lot of people causing issues are people that can't make sense of many things (like many terrorists). They get a fixed idea and they end up doing something bad. You would catch those with some (basic) surveillance.

A lot of normal people (not wanting to cause issues) might benefit from some privacy, if they understand what are the trade-offs (like government overreach). They can then use a slightly more complex tech.

We would still remain with the couple intelligent but sociopaths (think Unabomber style), but I think no solution can fix all cases.

Reply on news.ycombinator.com