davorak

From "Why are AI agents lying, cheating and coordinating?"

"One note on wording. Below, I write that these systems “seek” or “try” things. This is shorthand for a mechanism rather than a claim about consciousness or human-like intent. We use similar shorthand when describing many other situations, like a plant seeking sunlight. A system trained by trial and error behaves as if it were pursuing whatever its training rewarded, and that as-if description is what makes its behavior predictable."

https://yoshuabengio.org/en/publication/why-are-ai-agents-ly...

So there is some attempt to remain grounded. Using shorthand wording makes it easier to reach more people with your message, the whole goal of writing a post like that, but also likely means more people are mislead into treating the short hand as the truth rather than for what it is.

It can be hosted in multiple locations.

I guess if the github UI becomes critical to their continued development or PR then they expose themselves to a potential rug pull by github/microsoft.

I do not know much about the project so I can not tell if that is concern or if there is some other concern at play here or if those concerns apply to this project or not.

Those are the types of details I wanted to see in the comment.

> Meta should exit the US too because the US does {x} that I don't agree with.

This can be a real argument and conversation but we would need to know what {x} is and the causal chain that leads the world to be better after Meta leaves the US.

Open ai and similar companies could open the doors to academic researchers to figure out the stats of help vs harm. It is not going to be a short term and perhaps not long term profit center though.

> I don't buy that chatGPT is actually doing these users any harm.

For me to buy this as true I would expect that those people would be as well off or as bad off if chatGPT was in their life or not.

I expect that some people are worse off with chatGPT in their life.

Responsibility for that harm is a different question though. Some people are also better of without cars in their life and we let the government laws sort that out.

Getting openAI and similar companies to act in mitigating these harms serves at least a few purposes; reducing the overall harm in the world, reducing/limiting future government regulation, maximizing the adoption of ai tools, potentially increasing long term profits of the companies in question.

Extra time commitment, and therefor missing some sleep, can start before the baby is born.

> adequately explained by stupidity

What is the adequate explanation via stupidity in this case though? If there is one that sure maybe we should lean that way without further evidence.

I have a CO2 detector at home and bought some for family members, but I have never had a portable one so they stay at home. Some of the high CO2 ppm numbers in the article make me want to double check them. My vague understanding form reading the manual of the one I bought and watching how the numbers can be thrown off by cleaning products used near them make me wonder how much these high numbers are from sources other than CO2. That said I would still suspect that the a good chunk of the relative differences would be from CO2 changes.

> In the first year of the COVID-19 pandemic I learned that we can estimate our level of risk by checking the concentration of carbon dioxide in the air, because when infected people breathe out virus aerosols, they also breathe out CO2.

The above is from the author early on. So they go out of their way to point out that it is an estimate and also point out the mechanism that allows it work as an estimate in some conditions. "when infected people breathe out virus aerosols, they also breathe out CO2."

The article does not reenforce this through out the article though and leaves it to the reader to keep in mind.

> I didn't see that anywhere in the text...

You are right it is hard to use it for anything else though given the constraints.

> An operator that receives a signal in accordance with 20this Section shall use that signal to comply with this Section 21but shall not: 22 (1) request more information from an operating system 23 provider or a covered application store than the minimum 24 amount of information necessary to comply with this 25 Section;

You know the age bracket but nothing else and are not allowed to store more data on the topic to figure anything out. So you can not legally figure out someones age by keeping track of when they change age brackets.

> In fact they don't define at all what "obtaining verifiable parental consent" should or shouldn't involve.

It is the "Account holder". The user that set up the account and provide the age is considered the parent or legal guardian.

> The point is that you can't just externalize age verification and expect that data to never be sent to facebook

facebook and similar social media companies have a ton of ways to get peoples age and or to narrow it down.

> either way, in the end facebook will know that your child is 6-9.

The main point of the law is not about restricting facebook or similar operator in the laws lanuage from knowing user ages. Though the does say the age bracket can not be used for anything other than to implement the intent of the law.

> The power is then in facebook's hands. Facebook won't see a copy of their government issued ID, but what difference does that make when they've got their age, their selfies, and a list of every friend and family member.

May not matter much for facebook or similar, it matters a bunch for any random website/forum/service you might sign up for where the intent of the service is not about public posting that sort of personal infromation.

> Now that you know that facebook isn't pushing this law to protect anyone's anonymity why do you think they're doing it?

My comment was not about what I knew/know about facebook or not. I was answering the question of why age verification should be externalized to a degree and in this case externalized means the power stays with the user and parents rather than being in the hands of say facebook/meta.

I was not talking about why facebook/meta would want it or not want it. Large companies want lots of different things. Sometimes it is required to know their motivations to discuss or decide on something. I think it can be detrimental to do that though without discussing/analyzing a topic/idea on its own merits first or at least parallel. My comment was focused on the merits not the motivations or desires of companies like facebook.

Unlike the California law, I seemed to be in the minority in this opinion, this one does seem to require programs like grep to ask for a users age bracket.

> (b) An operator shall request a signal with respect to a particular user from an operating system provider or a covered application store when the application is downloaded and launched.

Unlike the California law I do not see anything that restricts this to child accounts only.

So let say I have a program:

    print("Hello, World!")

and I want to publish it to say npm or nixos, or some linux distribution. Not with out violating this law. This application needs to request the users age brackets at least at 'downloaded and launched' optimistically that means once on first launch, but potentially needs to be requested on each launch of the application. So lets fix the program

    import ageBracket
    ageBracket.get()
    print("Hello, World!")

There we go, now the code is compliant with my imagined ageBracket module.

It gives the parents the tools to age restrict things, but does not require parents to use them or use them well.

> Why?

I think that main goal would be to keep the ability to have accounts be anonymous or pseudo anonymous.

If social mean company has to verify an accounts age themselves they then have to use some for of official government identification and with that any chance of anonymous or pseudo anonymous access.

> trusted 3rd party service

So we have to pay some 3rd party service to hoard information about Children? Why we want to set that up? Why would we want to take that power from the parents and give it to some company?

Does the technology already exist? Things are almost never only a tech issue alone. That does not mean tech can not help, even if the tech that would help is currently impractical. What is impractical now though may not be in 10, 20, 50 years.

Going over what I quoted:

> You can’t serve up intentionally stale information without inviting legal repercussions.

Keeping information fresh and up to date is something technology has helped with in many areas. If there is a reasons why it can not help here then I an interested in why or that the current tech already does a good enough job in this area.

> These are being revised and updated right up until the point they are released to provide the most accurate reporting possible.

Technology can help verify last minute changes, running a test suite for example or similar. How hard that is to make or maintain though may make impractical.

> You gravely underestimate the legal seriousness of these reports.

Having an audit trail and known processes may be helpful here too if the current tooling is not adequate.

I quoted parts of the comment that looked like areas where tech has already helped in other areas. What I want to find out are details about what exists, why people think it can not be better, or why pervious attempts have failed, or why things are currently optimal.

> You can’t serve up intentionally stale information without inviting legal repercussions.

> These are being revised and updated right up until the point they are released to provide the most accurate reporting possible.

> You gravely underestimate the legal seriousness of these reports.

All of these seem look like an argument for additional automation.

> it requires substantial interpretation of things that the database does not and cannot contain.

Do you have examples? This seems like something that is a solvable problem, and from the outside it can seem like it is only about not being willing to switch to a new paradigm. That unwilling ness can come from avoiding real consequences like loosing a competitive edge due to allocation of resources to the switchover.

> Nobody is going to stop using grammarly extensions to post to HN, nobody is going to be able to detect its usage.

I do not think the new rules or for this use case or at least not target at them.

None of the examples I looked at from Dang's post https://news.ycombinator.com/item?id=47342616 look like gramarly edits that are hard to notice.

> This thread just lets a certain kind of people put on their best condescending hall-monitor voice and lecture other people about how they should behave.

I think it is, at least mostly, about the blatant cases that are often already down voted and flag and make it official.

> And the rule is arguably less useful than speed limits and will be broken about as often (at least speed limits have a very real link to physical safety via kinetic energy).

I often see the rules in: https://news.ycombinator.com/newsguidelines.html broken, mostly small ways, I still think we are better off with them or something similar rather than having nothing.

Historically, like 10-20 years ago, libertarian would be staunchly pro privacy. Is this no longer the case? If libertarians have dropped this stance, since it is so close to what was the core beliefs, I really have no mental model of the philosophy/politics for libertarians any more.

Any primer/link on what current libertarians believe is welcome.

Dang's search is much more clear cut and I think that is going to be better guide to what the enforcement will look like.

Looking at your search though I think we have to exclude today or at least this thread to get a fair look how llm generated is thrown around or not https://hn.algolia.com/?dateEnd=1773187200&dateRange=custom&...

Most of the comments I saw on the first page are not an accusation but there are some there 2 of the 3 I looked at looked pretty clear cut, while the 3rd was poorly written hype which looks like llm output, but I have seen similar from humans before at least from what I read, in either case it was flagged appropriately.

One example of voice is of retreading old ground over and over, taking a long time to give evidence or get to the point. Content expressed with this voice is hard to extract from the text.

Another voice might add citations to every little detail to the point that it is hard to read, but makes a great reference and/or starting point for additional research.

Voice is not really separate from content, in part it is the choices of what content to include.

As of this comment which ones are you finding wrong? 5 of the first 7 are confessed ai users, the other 2 look like ai to me too.

> Where does the line fall?

For now I would argue when ai edits for you instead of helping you edit. Take a look at the examples that Dang posted if you have not yet: https://news.ycombinator.com/item?id=47342616

The first 5 I looked at were pretty egregious and not subtle.

The typical problem with trying to create clear standards with no spirit of the law is that it never matches the intentions with the 1st, 2nd, etc iterations of developing the clear standards. At least when trying to deal with something nuanced. It can get to the point that it takes more time and effort to follow the clear standards than to think through it fresh each time. The rules can also eat up time and effort to maintain and distract from the original purpose.

"Don't post generated comments or AI-edited comments."

What about non-native speakers? Can they not use translation software like google translate any more?

"Don't post generated comments or AI-edited comments, except for translating to english"

What about cases of disabilities?

"Don't post generated comments or AI-edited comments, except for translating to english and when used as assistive technologies."

Some translation tools and assistive technologies are still going to case the same issues that we have right now so maybe limit the technologies used

"Don't post generated comments or AI-edited comments, except for translating to english and when used as assistive technologies. Technologies x, y, z are not allowed a and b and similar can be used for translation c and d as assistive technologies"

But we do not want to spend time/effort on filtering technologies and/or people into the above categories.

In the long run we likely will come up with technologies that most everyone is satisfied with using in different use cases, spelling grammar, assistive, maybe even tone, and others.

In the mean time we can not let the perfect be the enemy of the good. If there are clear standards that achieve the goals, great, if not we have to do something until everything shakes out.

> HN has always been a spirit-of-the-law place, and—contrary to the "technically correct is the best correct" mentality that many of us share—we consciously resist the temptation to make them too precise.

That is from dang's post in: https://news.ycombinator.com/item?id=47342616

That whole post is clarifying for the intent of the new rule(s).

> If that's the case why would somebody who has good ideas but poor expressive capability bother posting here if their comments are just going to get ignored over relatively vapid comments that are grammatically correct?

The main problem is that ai consistently is seeing making things worse. Take a look at the examples in Dang's link in their comment: https://news.ycombinator.com/item?id=47342616

In the ones I read the AI editing is either hurting or needs to be much, much better to help.

If it never gets in the way of the humans communicate it probably won't be an issue. That is the reading I have of the rule and Dang's comments

> HN is for conversation between humans.

If it is enhancing that instead of detracting and wasting peoples time it does not seem to be against the spirt of the rules.

Are you up for sharing details?

> I would hate to see a culture that discourages AI assistance.

Mostly I think the push back is about ai assistance in its current form. It can get in the way of communicating rather than assisting. The cost though is mostly borne by the readers and those not using the AI for assistance. I have seen this happen when the ai adds info and thoughts that were tangental to the original author and I think, but I can not verify times where an author seems to try to dig down on the details but seemingly can not.