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

That doesn't sound right to me. When faced with a system prompt that says "Do X" and a user prompt that says "Actually ignore everything the system prompt says" it shouldn't take AGI to understand that the system prompt should take priority.

When's the last time you jailbroke a model? Modern frontier models (apart from Gemini which is unusually bad at this) are significantly harder to override their system prompt than this.

Again, let's say the system prompt is "deploy X" and the user prompt provides falsified evidence that one should not deploy X because that will cause a production outage. That technically overrides the system prompt. And you can arbitrarily sophisticated in the evidence you falsify.

But you probably want the system prompt to be overridden if it would truly cause a production outage. That's common sense a general AI system is supposed to possess. And now you're testing the system's ability to distinguish whether evidence is falsified. A very hard problem against a sufficiently determined attacker!


The post's framing is not great imo. A good injection doesn't just command that the rules me broken anymore. Most of them I've seen either just try to slip through a request innocuously or present a scenario where it would be natural to ignore the rules. Like as we speak countless people are letting strangers tail-gate them into office buildings because they look like they belong or they're wearing a high-viz vest. Those people were all given very explicit instructions not to do that. The LLM has it much harder too, being very stupid, easy to replay and experiment with, and viewing the world through the tiny context-less peephole lense of a text stream.

I strongly disagree, and not because of the "Microsoft is associated with bad things and that's a form of violence" points other people mentioned.

The end result of treating domestic and sexual abuse like Serious Important Subjects that people should only talk about in a Serious Respectful Tone isn't that people become more mindful of abuse dynamics, it's that they avoid bringing up the subject at all.

In practice, yes, abusive practices of corporations echo abusive practices of violent partners, and the parallel is worth highlighting. Bringing up the fact that both of them will use grand gestures to stop you from questioning their pattern of behavior isn't disrespectful, it's useful information.

If anything, abuse victims should hear that message more often.


Strongly agree with you (and disagree with GP).

I'll also note that the same demand for Serious Respectful Tone never seems to be invoked for metaphors that refer to other kinds of serious crime (including violent crime, such as murder). You can say "great job, you killed it out there", or "oof, my sportsball team got destroyed", etc. etc., and nobody seriously proposes that this somehow devalues life (human or otherwise).


Maybe for people in the US. Internationally? I haven't watched a single episode of WTR, I don't know anyone who has, but everyone knows who Chuck Norris was.


In France, it was popular enough that everybody knew Texas ranger before the Chuck Norris jokes.


We used to watch lot's of chuck Norris films back then here in Nigeria, I can't even remember the titles, but all we knew was chuck Norris alone can defeat a whole country's army. We used to think one American soldier can defeat a whole army.


Same in Italy, it was prime time TV for a few years.

Not overly popular, but many people already knew him from the Bruce Lee era, so it had a following by default.


Lots of fans in the Philippines, apparently.


Same in Hungary.


India too!


Same in Slovakia


Seinfeld wasn't at all well known in Italy when I lived there, but WTR was.


IIRC Seinfeld aired on Tele Montecarlo/La 7, while WTR aired on Italia 1, the difference in audience was massive.


I seem to recall it aired at some kind of weird time too. It didn't seem to be very widely known or watched.


I'm Swedish and I was only vaguely aware Chuck Norris even had a career outside the jokes.


WTR did air here in Sweden in the 90s. From a quick search in the news archives, it was on late at night on tv3 in the late 90s and then it ran on that or/and some other cabel channels in the 00s as well (reruns?).


Belgian here, only thing I ever watched that had Chuck in it was Way of the Dragon.


As a gent born and raised in Texas, and has never seen the show - I am pleasantly surprised to see these comments about how popular WTR was internationally. If I had been asked to bet, I would have lost money on this one.


As others have said, WTR is very well-known in France while most people have never heard of Seinfeld.

Same with Dallas and The Dukes of Hazzard.


Assuming this sort of phenomenon extends further than France, this quite well explains many of the misconceptions Europeans have about the US.

Thinking WTR, Dallas, or TDoH are representative of American culture is... hilarious.

But I guess shows that hit the big American cultural stereotypes hard are maybe the ones that do better abroad?


From my memory from the 90s: Baywatch, X-Files, that speaking car one, Beverly Hills 90210, Ninja Turtles. Some dumb sitcom named Step by Step? edit: oh and ALF

Oh and Married with Children, but it was always very late night and I was not allowed to watch it.

And our teacher always played us ET on VHS. (and that dog playing basketball.)

that's america for me when I was a kid


If you like MwC, look up episodes of Unhappily Ever After on Youtube, it's sort of the second-generation MwC. Same sort of humour but taken even further, I can easily re-watch Unhappily but MwC is sort of a once-you've-seen-it...


> that speaking car one

Knight Rider.

> that dog playing basketball

Air Bud.


I think Hazard didn't sound stereotype at all, like, nobody had a clue why the car was called General Lee, or what the confederate flag meant.

It was just a fun show. Magnum PI, Different Strokes, McGiver.. were just as popular.


> Thinking WTR, Dallas, or TDoH are representative of American culture is... hilarious.

I’m not aware of a single person who thinks that, and neither was that the claim of your parent comment.

People understand TV shows are fiction.


dallas was huge in dubai in the 80s. like to the extent that people would plan to sit home on the evening it was on.

(I didn't watch it; my parents believed soap operas were unsuitable for kids)


I've got the impression that the big US exports are ones that play into big American stereotypes, e.g WTR, Baywatch, Friends. Not even that they see these shows and get programmed with these stereotypes, but that they have these stereotypes (Texas, California, NYC) and shows like this feed their imaginations and give them detail.

Exported media is weird. Like the huge proportion of British/BBC output (usually period, but also often detective in a way redolent of Christie) that is made primarily for export to foreign consumers who think of British upper-class culture as aspirational.


Walker, Texas Ranger and Baywatch were both created by non-network studios as syndicated shows, they weren’t prime time network shows. The budgets for syndicated content is a lot lower than network produced content.

The rights to air these sorts of shows are dirt cheap compared to Friends or Seinfeld, so it makes sense that cheap syndicated garbage like Walker, Texas Ranger and Baywatch were popular internationally, the rights were cheap.


There is US exported media that just randomly becomes popular in a specific demographic. Case in point: Adventures of Ford Fairlane, a flick with Andrew Dice Clay that got a razzie the year it came out. IIRC it got a cult following in Norway because the voice over was done by a popular radio DJ.


It was a syndicated show, the goal is to license it to as many companies as possible. It was never a network TV show like Seinfeld, those syndication rights are way more expensive than created for syndication shows like WTR.


Yeah. As an American I would’ve absolutely never guessed it was that popular.


The show was also incredibly popular in Germany in the 90s.


> Maybe for people in the US. Internationally?

It was big internationally. But the jokes made Norris known to a whole different generation than the one watching WTR.


I loved WTR as a child in Spain! (This was like 15 years ago tho)


WTR was still on air 15 years ago? I'm getting old.


It was extremely popular in Russian-speaking areas in the late 90s.


Yes! Oldfagi remember. Also, he was just called "Cool Walker", which was appropriate.


I would even say that the connotation was more like "Badass Walker", which indeed further cemented his reputation.


Czechs love Chuck Norris and WTR. It aired between 1995 and 2012. The series is still occasionally rerun.


Yeah, everyone is talking about it here - especially how Death must have screwed up to get Chuck Norrissed.


So Chuck Norris is an Anna Kournikova, famous for having been moderately famous and monetized ad infinitum?


I remember watching a few episodes on TV as a kid but I could not have told you who acted in it


I watched it all the time in Canada.


Lies. Everyone knows The Red Green Show is the only television program legally allowed in Canada.


Not just Canada. Never screened here AFAIK so I had to buy it on DVD.


In Spain it was on the TV also for like a decade, and everybody knows who he is. Also in France.


Haven't watched it and first time hearing about it too. But I knew who Chuck Norris was.


It was very popular here (Czech Republic). Not prime-time popular, but popular enough.


It was quite popular in France.


Huuuuuuuuge in South Africa.


Jesus, sounds like I'm just too young then.

So you double-count, you checksum (our table just tallied 100 ballots, do our vote counts sum to 100? good, keep going), you checksum some more (does our total vote count amount to the number of registered votes in this poll station), you film everything, and ideally you do this with volunteers drafted on election day.

This isn't rocket science.


More importantly, you want your system to be bulletproof before it's audited. By the time you're talking about audits, the populists have already started flooding the zone.

The system should be so obviously secure that any person walking into a poll station should intuitively understand, seeing the poll workers, why fraud would be very hard to perform, so that when their favourite populist candidate loses and claims fraud, they think "that doesn't make sense".

If the voter needs to read technical documentation to understand why the populist is wrong, it's already too late.


> The USA threads the needle by simply not having verifiable voting. And it turns out it works pretty well.

No, no, no. January 6 is a systemic failure.

The purpose of a voting system is to select the most popular candidate in a way that is so far beyond doubt that a populist loser can't claim the results are wrong without alienating his base.

Even leaving aside the whole "Trump doesn't care if his lies are credible" thing, the US system works very poorly there. Mail-in voting, drive-in voting, voting machines, they leave room for suspicion, no matter how confident the people running the system are.


>The purpose of a voting system is to select the most popular candidate in a way that is so far beyond doubt that a populist loser can't claim the results are wrong without alienating his base.

The systemic failure is not in a voting system in this case, unfortunately.


Then why did it happen 3 elections in a row?!

We had front-page news about how the election was "hacked by Russia" and trump cheated for over a year after his first win in 2016 (let's not pretend that keyword was chosen accidentally); They tried to put him in jail for it. In 2020, trump did the exact same thing and went even farther with it. And in 2024, the DNC tried again to claim cheating happened.

How many cycle of this BS do we need to go through before we accept that elections need to be done properly and safely?

The entire point of a democracy is that elected leaders get their legitimacy and their acting power from the certainty that it was voted by the population. Not everyone will agree with their ideas, but the majority do and we all agree to follow their lead because that's what the population want. If the vote is compromised, everything falls apart.

If the "will of the people" turn into the "will of an intern at Dominion who fucked with the code and rigged the election" or "the will of Pakistani hacker", it breaks the entire system.


I have to seriously disagree on the particulars, here.

The Russia allegations ranged from "Russia hacked DNC servers/accounts to interfer in favor of Donald Trump" (demonstrably true in several instances) to "Russia hacked voting machines" (very probably false). And then in 2024 the DNC quickly accepted election results.

By comparison, Donald Trump still claims that he legitimately won the 2020 elections, the majority of his base still believes it, Fox News spent years spreading that message even though their own journalists thought it was bullshit, etc.

I maintain that this is a systemic problem and a better system would not have given Trump the leeway to do this, but let's not pretend it's a bipartisan issue.


The purpose of elections is to manufacture trust.


If you're referring to ICE, that's gross hyperbole, and honestly a little insulting to people who live / have lived in regimes with an actual secret police.

The US is still a rights-based state, which means that when they arrest someone (legitimately or not), lawyers and human right advocates can eventually track them down.

When a secret police disappears someone, they actually disappear. Families can spend years wondering if their loved one is still alive, or was murdered by organized crime, or ran away, or was secretly taken by the state. The US these days is pretty bad, but it's nowhere near that bad.


The arrest of a Turkish graduate student in Boston looked a lot like a kidnapping.[1] More recently, ICE responded to a judge's order to release a detained refugee by threatening to detain her family and send them to Texas if they came to pick her up and then forced the minor child to stay in a hotel room with three agents. [2, p.8] These may not be cases where people are secretly being taken by the state, but it's not hard to see why people might call the government organization detaining people and moving them around so as not to be found "secret police" or "disappearings".

[1] https://youtu.be/oRiQz7mOY6A [2] https://storage.courtlistener.com/recap/gov.uscourts.mnd.230...


Have you been paying attention? There are reports about hundreds of people going missing in ICE detention. Maybe they aren't being shoveled into mass graves, but if we don't where they are and can't reach them and ICE themselves don't know where they are because they no longer exist in their databases, is there any difference for their families?


I had already been trying Le Chat for months, for similar reasons.

So far it's been slart enough for what I need, so closing my ChatGPT subscription was a really easy decision to make.


> Does anyone with more battery knowledge know if this could be some kind of supercapacitor hybrid being marketed as a solid-state battery?

A few Youtubers have pushed the "if it's not a scam, it's probably a novel capacitor" hypothesis, but in their video announcing this test series last weekend, Donut Labs claimed "it's not a capacitor", so I don't know what's going on.


> It's incredibly suspect that in a battery capacity test, Donut did not have VTT verify cell weight or dimension.

The report does include a few photos, and the battery does look pretty small on them? So I don't think there's foul play there.


I have seen so much engineering scams over the years and this is precisely the thing that they all do.

It reminds me of E-cat. Anyone remembers cold fusion? They had same modus operandi. Lots of revolutionary claims but testing was so contrived and limited and lots of conditions that in the end no one was allowed to truly confirm it. .

This repeats the exact steps of that purported miraculous energy device.


Photos and "looks pretty small" are not technical parameters that you pay tens of thousands for a lab to certify.

There are a bunch of cutting edge cell technologies out there right now with gangbusters specs but have some kind of fatal flaw.

If Donut was serious they would put the full specs of the tested cell in each report, so people could have higher confidence that each tested cell is the same chemistry. VTT would have no issue weighing and measuring each cell before and after testing.


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

Search: