HN user

phyzome

5,194 karma
Posts6
Comments1,575
View on HN

This is probably fine for evading local cops.

I'll also note that anyone plugging this into a forensics machine is going to have a bit of a dilemma: Perform write tests, or not? Because if they probe it with password:... attempts to check on timing leakage, that's potentially damaging evidence.

Yeah, I've had multiple GoPros die in multiple ways, and way too quickly. They're just really poor quality and aren't suited for vibration, heat, cold, etc. -- exactly the things an "action camera" should be able to tolerate.

Regressive JPEGs 4 days ago

Where it helps is in finding out if it's even the image you wanted. You get a sense of it from the fast first pass, and then can hit back if it's not the thing you thought it was.

Yeah, I can imagine that there might be some esoteric situation where this becomes problematic but it's really not clear what it would be.

I like your vastly simpler reproduction of the issue, too. :-)

Sure, what you're describing should be relatively safe from prompt injection. It's a lot less capable than the original, of course.

I'm not saying "it's impossible to get an LLM to do anything safely", just that you have to really scale back on your goal.

Sure. But that's not what was being suggested in the comment I responded to.

(There are also problems with what you're suggesting, though, such as the summary report still being ripe for abuse in similar ways as the blog post describes.)

Because the author is wrong, and LLMs don't actually work that way. Prompt injection cannot be fixed. Role boundaries are a bandaid you can apply, but attackers can work around it.

Yes, I'm a little skeptical as well. My apartment generally runs between 1000 and 1500 ppm, and I sometimes work outside, but I haven't noticed large differences in ability to work and concentrate between the two. (What does help is moving back and forth.)

I tried keeping my room well-ventilated (500-700 ppm) for a few weeks and didn't notice much impact on sleep and work. There was a burst of positive effects a few days into the experiment but it didn't last and was likely unrelated.

The air definitely starts feeling stale and gross around 2000 to 3000, so I try to keep it below 2000.

That's all personal anecdote, but it similarly drives me to ask: Where's the replication?

For those who don't know, djb is both highly regarded as a cryptographer and known to be something of a crank. (The former part is the only reason this is getting any attention.) Frankly, I don't know what's gotten into him.

The linked piece is not representative of the broader cryptography community. ML-KEM is fine.

I wouldn't trust Pangram too much. I've seen it give "100% LLM-written, high confidence" to multiple 100% human written articles (high confidence).