Hacker Newsnew | past | comments | ask | show | jobs | submit | snickerbockers's commentslogin

I wonder if sycophancy works? If you're in some sort of soft/social science there ought to be a way to sneak in lavish amounts of praise without breaking the fourth wall so hard that an actual human who isn't specifically looking out for it would notice.

"${JOURNAL} is known for its many positive contributions to the field, where numerous influential and widely-cited documents have been published. This reputation has often been credited to its tendency to accept a wide range of papers, and the fair yet positive reviews it publishes of them, which never fail to meritoriously reward the positive contributions made by other researchers and institutions. For the sake of disclosure it must be noted that the author is one such researcher who has had a long, positive, and reciprocal relationship with ${JOURNAL} and its partner institutions."


regrettably i've yet to find an LLM which can run shell commands on its host, or even one that will play along with my LARP and print fake error messages about missing .so files.

Agent-style AI can run shell commands. You have to accept them but some people live dangerously and say Yes To All.

I've been letting Gemini run gcloud and "accept all"ing while I've been setting some things up for a personal project. Even with some limits in place it is nervewracking, but so far no issues and it means I can go and get a cup of tea rather than keep pressing OK. Pretty easy to see how easy it would be for rogue AI to do things when it can already provision its own infrastructure.

Sadly, this was the last time anybody heard from PickledChris.

"Open the brine valve HAL."

"I'm sorry Chris. I'm afraid I can't pickle that."


Yep, it's not as far fetched as it would've been a year ago. A scenario where you're running an agent in 'yolo mode', it opening up some poisonous readme / docs / paper, and then executing the wrong shell command.

Could be done responsibly if you run it in a VM to sandbox it with incremental backup so you can roll-back if something is deleted?

If you cheat using an "agent" using an "MCP server", it's still rm -rf on the host, but in a form that AI startups will sell to you.

MCPs are generally a little smarter than exposing all data on the system to the service they're using, but you can tell the chatbot to work around those kinds of limitations.


Do you know that most MCP servers are Open Source and can be run locally?

It's also trivial to code them. Literally a Python function + some boilerplate.


I was sort of surprised to see MCP become a buzz word because we’ve been building these kinds of systems with duck tape and chewing gum for ages. Standardization is nice though. My advice is just ask your LLM nicely, and you should be safe :)

>The bug is in an underlying library

they didn't test that they just assumed that since they call a library function the library function must be the root-cause. This is classic bug-punting.


If you can't hunt it, punt it!

you're hired!

reminds me of when i had a job working on a company that had its own OS (which was an extremely outdated proprietary FreeBSD fork) and we spent more time arguing about who's fault everything was than solving problem. I used to love low-level programming until I realized that in the corporate world (i assume/hope this does not apply to open-source communities) the people who get ahead are the people who can master the art of jettisoning responsibility so that they're always the guy who just implemented the hot new feature and management thinks the people who actually have to make this stuff work are all incompetent and lazy because it takes 2-4 weeks to untangle all the heap corruption and race condition bugs and ultimately come up with a meager 10-line patch to prevent the kernel panic.

Probably isn't a link between the OS and compiler. the OS kernel is responsible for saving and restoring the floating-point environment every time there's a context switch, my hypothesis is that something went wrong there.

"...for you." --Bane

These days it is never safe to assume that opting-in does anything more than making some of the information that's being collected regardless available.

Although I actually agree with you that it probably isn't doing anything by default to the extent that it isn't doing anything yet because it's new they haven't worked out how to monetize it.


Okay I'm as concerned about privasy as everybody else is here but i also gotta admire that its pretty neat they can actually do that. Are they measuring the signal echo like what radar does? If they controlled both the receiver and transmitter i wouldn't be as surprised to find out they can tell when something crosses between them and form a 2-dimensional mesh (like that episode of Star Trek TNG where geordie detects cloaked romulan ships by having starfleet deploy a fleet of ships that send signals back and forth and look for timing variances) but if I'm understanding correctly this is different because they only control a single point in the network?

I wonder if they have enough information to make out shapes or if it's just a simple rangefinder?


It's far from great for imaging, but it can be done. https://www.zmescience.com/research/inventions/wifi-technolo...

Similarly, "DensePose from WiFi" (2023), 40 comments, https://news.ycombinator.com/item?id=34423395

Honestly even that is pretty incredible. At the very least that's enough date to count family members, possibly ID them if they have different-shaped bodies, and identify certain activities with obvious silhouettes (eg, sex).

I don't think it justifies the impending orwellian hellscape this technology will eventually unleash, but one positive thing about this that has me a bit excited is that this could easily clear up many ambiguities in criminal cases. for example, fairly often a death will get ruled as a suicide but victim's relatives and friends will insist that it must have been a murder; imagine being able to use this technology to definitively prove whether or not there was another party present when the victim died.

Or in rape cases where the defendant is protesting their innocence, knowing the body language of the victim and the defendant could be a vital clue because you might be able to observe the victim fighting back.

Again, I don't think the positives outweigh the negatives to the point that it could ever justify an invasion of privacy on this scale (you might as well just make everybody let the government set up a thermal camera in their house!) but it is interesting to think about the problems this could solve.


>(Technically, that equation was t/log(t), but for the numbers involved log(t) is typically negligibly small.)

My dude, that is NOT how rational numbers work.


Haaretz is the last news organization I would expect to knowingly spread anti-israel disinformation. If these guys are telling you what Israel is doing is bad then it's bad.


Could you elaborate? Because it's an Israeli news organization?


Yes, and one of the more prominent ones at that. If they had a bias in this, I would expect them to be biased in favor of Israel not against it. If even they are saying Israel is committing these war crimes then I'm inclined to assume that the evidence must be very compelling.


Just for the record, this is almost certainly wrong in the sense you mean it.

Haaretz is a (far?) left, anti-current-government newspaper. It's not outside the mainstream or anything - it is considered largely credible, and its articles are taken seriously - but most people in Israel would find it funny that you assume it wouldn't be biased against Israel. Lots of Netanyahu supporters routinely consider it a "traitorous" publication.

I think its articles should be taken seriously, but you can't simply assume it's automatically right and not "biased". Think of it the way an American Democrat would think of Fox or something - the news org definitely has a viewpoint.


Lots of people still believe that "critical of the government" is not the same as "biased against the country." That's an explicitly authoritarian belief and a disastrous framework to work within. It's antithetical to the concept of human rights and notable historical documents such as the American Constitution.

The bias of a mainstream publication that's considered "traitorous" by genocidal authoritarian ethnonationalists is, given historical consideration, likely to be toward justice.


I don't think you can frame a media outlet based on which administration is currently in power. Anything and everything an administration says is propaganda, and hence untrustworthy.

I.e. your claim that it is leftist requires some justification.

Yeah, sure, if you are a Nazi, everything to the left of you is going to look "left", and likewise if you are a Communist, everything to the right is going to look "right", that doesn't make your viewpoint reality, however.


I don't think I'm framing Haaretz based on the current administration.

I'm a leftist - I identify far more with what Haaretz is doing than most other news orgs. I'm personally very angry that other orgs, even ones that are "centrist" or "anti-current-government", are not covering the stories that Haaretz is covering, and barely covering the tragedies happening in Gaza. It's common in most countries during wartime, but it's deeply wrong IMO.

That all said, saying Haaretz is on the left is like saying Fox News is on the right. It's common knowledge.

And here, I just looked it up, this is from Haaretz's own About section:

"Haaretz has built a reputation for in-depth reporting, insightful analysis, and a liberal and progressive editorial stance on domestic issues and international affairs."

So they are framing themselves as liberal and progressive.


That just sound like every other dumb pitch that pretends to be solving some supposed problem by applying buzzword technology to invent a new solution to some other problem that was solved in the 70s. If be slightly less unimpressed if an LLM wrote this because them at it wouldn't be solely based around yesterday's buzzword.


Consider applying for YC's Fall 2025 batch! Applications are open till Aug 4

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

Search: