lolinder

> In order for the LLM to emit sensitive data publicly, you yourself need to explicitly tell the LLM to evaluate arbitrary third-party input directly,

This is the line that is not true.

When it comes to security a threat actor is often someone you invited in who exceeds their expected authorization and takes harmful action they weren't supposed to be able to do. They're still an attacker from the perspective of a security team looking to build a security model, even though they were invited into the system.

> There is no attacker in this situation. In order for the LLM to emit sensitive data publicly, you yourself need to explicitly tell the LLM to evaluate arbitrary third-party input directly,

This is not true. One of the biggest headlines of the week is that Claude 4 will attempt to use the tools you've given it to contact the press or government agencies if it thinks you're behaving illegally.

The model itself is the threat actor, no other attacker is necessary.

> Any time you use an LLM with tools that might be exposed to malicious instructions from attackers (e.g. reading issues in a public repo, looking in your email inbox etc) you need to assume that an attacker could trigger ANY of the tools available to the LLM.

I think we need to go a step further: an LLM should always be treated as a potential adversary in its own right and sandboxed accordingly. It's even worse than a library of deterministic code pulled from a registry (which are already dangerous), it's a non-deterministic statistical machine trained on the contents of the entire internet whose behavior even its creators have been unable to fully explain and predict. See Claude 4 and its drive to report unethical behavior.

In your trifecta, exposure to malicious instructions should be treated as a given for any model of any kind just by virtue of the unknown training data, which leaves only one relevant question: can a malicious actor screw you over given the tools you've provided this model?

Access to private data and ability to exfiltrate is definitely a lethal combination, but so his ability to execute untrusted code, among other things. From a security perspective agentic AI turns each of our machines into a Codepen instance, with all the security concerns that entails.

Exactly. Turning tertiary education into a third tier of babysitting just screws over the adults who actually grew up during secondary school. Tell them how to succeed in your class and then let them fail if they won't listen to you! It's high time someone let these kids grow up.

I'm not questioning that NYC's subway is a disaster. Yes, I've spent time in it and have spent time in a few specific European subways and am well aware of the difference.

I'm all for questioning why NYC's subway is worse than London's. I just think that questioning why "cities in Europe and China can afford all this stuff and we (Americans) can’t" is a bad framing: it's so general as to be meaningless and perpetuates misleading stereotypes about both sides of the pond.

The fact at hand is that NYC specifically is a deeply corrupt and extremely filthy city by any standard, European or American.

This is a pretty sweeping generalization both on the Europe+China side and the US side.

I'm certain that there are cities in Europe and China that struggle just as much with corruption and graft as NYC does and there are also cities in the US that have substantially less corruption and graft than NYC. It's just that it's vogue right now to cherry pick American problems and European/Chinese successes rather than the other way around.

Is it connected with an organization? I don't see any evidence of that in the About page or anywhere else. The donations page says to find a local wildlife sanctuary and donate to that, then links out to two options if you really can't find one of your own. But I see no evidence that it's associated with either one of those entities it links to.

https://owlsintowels.org/support/donate/

All of the best professors I had either did not grade homework or weighted it very small and often on a did-you-do-it-at-all basis and did not grade attendance at all. They provided lectures and assignments as a means to learn the material and then graded you based on your performance in proctored exams taken either in class or at the university testing center.

For most subjects at the university level graded homework (and graded attendance) has always struck me as somewhat condescending and coddling. Either it serves to pad out grades for students who aren't truly learning the material or it serves to force adult students to follow specific learning strategies that the professor thinks are best rather than giving them the flexibility they deserve as grown adults.

Give students the flexibility to learn however they think is best and then find ways to measure what they've actually learned in environments where cheating is impossible. Cracking down on cheating at homework assignments is just patching over a teaching strategy that has outgrown its usefulness.

What you're missing is that it's right on the cusp of being a learning tool instead of a silly game because it has regressed below the cusp. It was a useful learning tool, and its trajectory has been strictly downward for years now.

— Sir Arthur Stanley Eddington, The Nature of the Physical World (1927)

Agreed. Duolingo started out on the right foot: they had gamification, but not too much of it, and they clearly cared about helping you learn. For a long time it was the most highly recommended app for learning new languages and that wasn't just naivete, it actually did work.

That's changed gradually over the last few years as they switched from using gamification in pursuit of learning to using a veneer of learning as a pitch to get people to try their game.

My wife quit Duolingo the week before this announcement after years of watching Duolingo prioritize attention manipulation over learning. She had a nearly 6-year streak and was on the paid version at the time, but realized that it wasn't actually helping her learn any more: she'd at some point begun maintaining a streak just for the sake of maintaining a streak.

The best documentation for Duolingo's decline is this article from a few years ago [0]. It's a piece by Duolingo's CPO (who was a former Zynga employee) where he discusses at length how Duolingo started using streaks and other gamification techniques to optimize their numbers. He has a lot to say about manipulating users into spending more time with them, but in the entire piece he barely even gives a token nod to the supposed mission of the company to help people learn. The date he cites for the beginning of their efforts to optimize numbers pretty closely correlates to my sense for when my wife began to complain about Duolingo feeling more and more manipulative and less and less useful.

This past month they finally jumped the shark and she decided to quit after 6+ years. The subsequent announcement that they'd be using AI to churn out even more lackluster content gave us a good laugh but was hardly surprising: they'd given up on prioritizing learning a long while ago.

[0] https://news.ycombinator.com/item?id=34977435

> (1)

They're natural, sure, but not really natural in a way that directly affects the most common everyday use case for temperature: the weather.

Fahrenheit's 0-100 is for now approximately the range of "normal" weather for the continental US, which is a useful if accidental property for Fahrenheit as a weather temperature system. Celsius's use of water's boiling point as a value for 100 is a nice property from an aesthetic perspective but means that it doesn't use the full 100 degrees when representing any earth-based climate.

> (2) ... (3)

Both of these aren't really advantages if you treat the scales as arbitrary, which in either case you are for most real-world uses of either scale. Water's boiling point at sea level is totally irrelevant when you're measuring the temperature of the outdoor air, a human body, or a steak.

> (4)

Only matters if you're actually doing work that needs to translate to and from Kelvin, which most people will never do in their entire lives.

> (5)

Standardizing would be helpful for sure, but it's not obvious that it would be worth the headache of making the switch given that the only real advantage to Celsius for most use cases is the mere fact that it's a standard.

Both processes create/release heat, and in Alec's tests in the linked video the bulk of the heat from the dehumidifier running in a humid space is coming from latent heat released from the water. That may not be true in regular operating conditions, but there will always be a substantial amount of heat released from the water when you trigger a phase shift.

> I don't think there has to be any heat involved with collecting water molecules in the air into a larger volume of water, depending on the process used.

The only other option is to increase the pressure in the room or in a space within the room, which this material pretty clearly isn't doing.

Yeah, there's certainly some of that, but going off of Alec's numbers and trial above you end up with a larger portion coming from the latent heat than from the mechanism, for the same reason that a heat pump is more efficient than an electric heater: it's condensing more water than it is running mechanisms.

That said, his demo is not under typical operating conditions in that a dehumidifier is normally expected to actually be able to catch up and reduce humidity in the room, while his demo ensures that the humidity levels stay high throughout the hour. So it's likely that under normal operating conditions the mechanism's proportion of the waste heat is higher than it is in his demo.

I totally understand that inversion but I think it's a bad analogy.

Industrial automation works by taking a rigorously specified designs developed by engineers and combining it with rigorous quality control processes to ensure the inputs and outputs remains within tolerances. You first have to have a rigorous spec, then you can design a process for manufacturing a lot of widgets while checking 1 out of every 100 of them for their tolerances.

You can only get away with not measuring a given angle on widget #13525 because you're producing many copies of exactly the same thing and you measured that angle on widget #13500 and widget #13400 and so on and the variance in your sampled widgets is within the tolerances specified by the engineer who designed the widget.

There's no equivalent to the design stage or to the QC stage in the vibe-coding process advocated for by the person quoted above.

Less energy would definitely be a huge plus but unless this violates our understanding of thermodynamics there will still be unwanted heat put out into the air. The heat from a dehumidifier comes primarily from the latent heat in the water being released so that the water can become liquid. This heat must be released somehow in this process unless they actually did find something that breaks our understanding of physics.

Obligatory Technology Connections video on the topic: https://www.youtube.com/watch?v=j_QfX0SYCE8

Why are you framing this as being primarily about Google being un-sueable? There's clearly a problem with Google being difficult to work with to re-claim ownership of a business profile (no customer support, as always) and Google obviously has deep pockets that would be tempting to get access to, but isn't Lieferando the one engaged in the extortionate business practices?

Under US law I can see a few different things that would make the Lieferando behavior you describe illegal, whereas all Google is doing is being the unwitting vector for their illegal activity.

It's always more difficult to pin fault for a crime on unwitting enablers even when their negligence arguably rises to the level of a crime. The big question here is why businesses haven't successfully fought back against the ones doing the actual crime?

Google should be forced to do this, but also I'm pretty sure that this behavior by Lieferando breaks existing laws that just need to be enforced.

OP did not market it as universal, they explicitly say it is an iOS app in the description of the Show HN and on the website.

Honestly, your initial comment comes off as whiny and entitled: you lost maybe 30 seconds of your time clicking through to something that wasn't relevant to you, and there's "no excuse" for leaving a drive-by comment attacking the author for their choice of tech stack for a side project just because of that tiny inconvenience.

Garmin was a long-standing well-respected consumer hardware company that started releasing a miniaturized version of the thing they were famous for: a GPS. They already had brand recognition and all the infrastructure to build quality products that people would trust.

That infrastructure and brand is extremely difficult to bootstrap.

Also, you know who did measure every angle to make sure it was correct? The engineers who put together the initial design. They sure as hell took their time getting every detail of the design right before it ever made it to the assembly line.

Who's filling that role in this brave new world?

But the point is that you're going to have a hard time bootstrapping a new phone ecosystem in the current environment. If it's mission critical then they need to be throwing everything at it, not doing it as one option among many.

When I think of OpenAI hardware I can't help but think of Akins Laws of Spacecraft Design [0]:

> 39. Any exploration program which "just happens" to include a new launch vehicle is, de facto, a launch vehicle program.

Having a Jony Ive project on the side isn't going to do squat for OpenAI—if they're going to go into consumer hardware that's going to need to be an all-consuming strategic pivot, which their other moves suggest they're not doing. They're currently in spaghetti at the wall mode with Jony Ive as just one bet among many, which is a very bad way to approach a new piece of consumer hardware that's meant to compete with Apple.

[0] https://spacecraft.ssl.umd.edu/akins_laws.html

The slope of the scales has no bearing on whether percentages are meaningful here. The problem with both systems when it comes to percentages is that neither system has 0 set to a natural 0. This leads to an entirely arbitrary point on the scale where decreases in the unit will approach a 100% difference and then suddenly start decreasing again.

If anything Fahrenheit should be less insane because at least the artificial 0 is likely to stay much further away in the data they're quantifying so the percentages stay reasonable.

It can be a science fiction result and still not actually result in time saved for the human operator on the whole. For me the jury is definitely still out on whether it results in net time saved and it's not for lack of trying.

Whether it ends up getting good enough in the near future that it does become a net positive both isn't the question being discussed and still remains to be seen.

They didn't want it to be a website, and if they did make it a website they'd have to deal with the inevitable entitled whining on their Show HN wondering why everything has to be a website these days.

If there's one thing I've learned from a few decades on the internet it's that the only way to avoid entitled whining is to never, ever release anything, especially not for free.

> in the last few months after recently becoming a father ... I decided to build an iOS App for me and others ... I uploaded the app on the app store for free and no ads

OP wrote an app to solve their own problem and made it available for free for others to use if they find it helpful. It should shock no one and certainly should not trigger any judgement or attacks from anyone that they chose to write the app for the platform which they themselves use.

When used that way they also regularly get into loops where they lose track of what they were supposed to do.

The last time I set Cursor on something without watching it very very closely it spun for a while fixing tests and when it finally stopped and I looked what it had done it had coded special cases in to pass the specific failing tests in a way that didn't generalize at all to the actual problem. Another recent time I had to pull the plug on it installing a bunch of brand new dependencies that it decided would somehow fix the failing tests. It had some kind of complete rewrite planned.

Claude Code is even worse when it gets into this mode because it'll do something totally absurd like that and then at the end you have to `git reset` and you're also on the hook for the $5 of tokens that it managed to spend in 5 minutes.

I still find them useful, but it takes a lot of practice to figure out when they'll be useful and when they'll be a total waste of time.