Hacker News new | past | comments | ask | show | jobs | submit | pelario's comments login

Despite the appearance, they do: despite the training, neurons, transformers and all, ultimately it is a program running in a turing machine.

But it is only a program computing numbers. The code itself has nothing to do with the reasoning capabilities of the model.

Nothing to do with it? You certainly don’t mean that. The software running an LLM is causally involved.

Perhaps you can explain your point in a different way?

Related: would you claim that the physics of neurons has nothing to do with human intelligence? Certainly not.

You might be hinting at something else: perhaps different levels of explanation and/or prediction. These topics are covered extensively by many thinkers.

Such levels of explanation are constructs used by agents to make sense of phenomena. These explanations are not causal; they are interpretative.


> Nothing to do with it? You certainly don’t mean that. The software running an LLM is causally involved.

Not in the way that would apply problem of non-computability of Turing machine.

> Perhaps you can explain your point in a different way?

LLM is not a logic program finding perfect solution to a problem, it's a statistical model to find next possible word. The model code does not solve a (let's say) NP problem to find solution to a puzzle, the only thing is doing is finding next best possible word through statistical models built on top of neural networks.

This is why I think Gödel's theorem doesn't apply here, as the LLM does not encode strict and correct logical or mathematical theorem, that would be incomplete.

> Related: would you claim that the physics of neurons has nothing to do with human intelligence? Certainly not.

I agree with you, though I had different angle in mind.

> You might be hinting at something else: perhaps different levels of explanation and/or prediction. These topics are covered extensively by many thinkers. > Such levels of explanation are constructs used by agents to make sense of phenomena. These explanations are not causal; they are interpretative.

Thank you, that's food for thought.


Well, if you break everything down to the lowest level of how the brain works, then so do humans. But I think there's a relevant higher level of abstraction in which it isn't -- it's probabilistic and as much intuition as anything else.

> Power it off

This has been the real game changer for me. As instant gratification is not so instant, I can go way larger stretches of time without checking the phone. Also, as nowadays most people message, and calls are not so common (at least in my circles), there is no "harm".


Of course there are tons of open questions, yet you made a huge false dichotomy.

I would bet that most people agree with banning sales of alcohol, cigarettes and opioid to kids.


But it would be silly to classify opioid use as gambling, which was the proposal. Figuring out the real issue, and banning that for kids, might well be a good idea. But to my mind the thinking from parents is still very open around what kids should be allowed to do, so it might be too soon.


The argument was to treat it like gambling.

Not to classify it as gambling.


How about Wikipedia's approach?

Of course Wikipedia is way smaller than the internet, but still one way to go could be by having themed "human curated niches"


Are you saying that none of their original appliances have died?

If not, do you realize that you claiming " If it was survivorship bias some of these appliances would've died." When "these" refers exclusively to those who haven't died is the very definition of survivorship bias?


I think he's saying Mom would have definitely remembered if she had had to replace every darn appliance in her house every 2-5 years, the way we do now, before eventually lucking out to find the one good washer, dryer, dishwasher, etc. made in their respective years.


I think a washer died around 35 years ago and her current washer is 35 years old. It seems improbable that so many 40+ year old appliances survived in one home if reliability rates or ease of repair were not significantly better in the past.


Except 13×28 = 364, so you get the "free day" every year, and two of them on leap ones


I don't know how common is that story... Either I know you, or I know someone with a similar story.

Anyways, as someone who identifies both as a computer scientist and as a dancer, I can definitely relate to your experience:-)


That's really an example of bad journalism. What you find in the text is quite distinct from the title:

> Hawking announced: “I have changed my mind. My book, A Brief History of Time, is written from the wrong perspective.”


Bad headline-writing, rather.


Flag clickbait titles.


What about phone companies ?


Seems global outage: https://www.cnbc.com/2022/10/25/whatsapp-is-down-globally.ht...

Each time this happens I try to recruit more people to use Signal.


How is Signal any better? US based, centralized, hostile towards alternate clients. Not much better IMO.


US based isn't a concern in this case, as the data they store is pretty much unusable.

They publish the subpoenas they receive and the dump of relevant data to the authorities, and it's usually a timestamp associated with some ID and that's it.

https://signal.org/blog/looking-back-as-the-world-moves-forw...


Actually very nice to integrate with. I'm not using Google or Apple in my life. On the phone I'm using sailfish OS, so the main stream apps are not usually ported natively. Fortunately someone used libsignal and added frontend so signal is my main means of communication with friends. And I still don't have to drown into Google or FB services.


Much better in that it is open source (so you can audit the e2ee), and it does a lot about metadata. It is actually better at protecting metadata than many decentralized alternatives.

Nobody said it was perfect. It's just better.


WhatsApp actually uses signal protocol for e2ee


They say they do, but you have to trust them. You don't have to trust Signal, you can audit the sources (or trust some third party to do it for you).


How do you know the source code you’re looking at is for the same program you downloaded from the App Store? Does apple publish a checksum of software you’re installing?


> Does apple publish a checksum of software you’re installing?

Reproducible builds: https://github.com/signalapp/Signal-Android/tree/main/reprod...


This is a failure of Apple and their walled garden, not of Signal. If this is a concern to you, you either need to jailbreak or switch to a more free as in freedom platform.


Honestly, if it really mattered a lot to me (i.e. to my own security), I would compile Signal from source and install it on my device. Which I could not do with WhatsApp.


True


Signal doesn't press you to setup (by default) unencrypted cloud backups.


It doesn't matter, the app is closed source so they can still access your messages regadless of what protocol they use.

https://gizmodo.com/whatsapp-moderators-can-read-your-messag...


That article doesn't support what you're saying. It says that WhatsApp has access to metadata, which it hands over to law enforcement. This does not necessarily mean that they can read your messages.

It does say that 'WhatsApp can read some of your messages if the recipient reports them'. That 'if' is doing a lot of work in this sentence. It means that the recipient has to decrypt your message.

Although there are forms etc. within the app for doing this, it's essentially no different to taking screenshots.

There is no way to ensure 100% privacy if the other party you are communicating with does not keep data they have access to private.

I'm not a big Meta fan, but as far as I am aware, they can't normally read your messages. The fact that it's closed source just means that we can't verify that for ourselves.


> The fact that it's closed source just means that we can't verify that for ourselves.

That's really the whole point. As far as we know, it could be that it is not e2ee at all.

Also from the moderation article, it's not clear to me what that means: if I report you, does that mean that the moderators will get access to all your recent conversations? Could be, right? But then the FBI could report you for no reason, and then ask WhatsApp to provide your recent conversations. Which would effectively act as some kind of backdoor, right?


I agree, closed source means we can't do anything apart from decide whether we believe Meta or not.

But my understanding is that the 'report' is from a user's WhatsApp client—if someone sends you a message that you think is reportable, you can report them to Meta. As part of the report, your WhatsApp client will forward some information to Meta.

Assuming Meta are not actively lying, this would not mean that it's not E2EE.


Can I see the source to make sure?


I'd say it's a lot better since it doesn't do unknown things with your address book. It actually doesn't do any unknown things and the fact that they're US based is irrelevant since they have nothing to give away thanks to E2EE.


What kind of unknown things with an address book is WhatsApp doing?


The idea is that Meta uses WhatsApp contact lists to build and maintain their social graphs, which is in turn used for advertising.


WhatsApp uploads the address book from the user's phone to Meta's servers and after that it's unclear to outsiders what they do with it. Hence "unknown".


Is this not just another centralized point of failure?


I'm presuming GP does not put Signal forward as a solution to outages. But instead uses the opportunity to talk about messengers and show that there are alternatives in general.

At least, that is what I do.


Yes it's what I do, and then I actively struggle to hold back on pushing Matrix/Element as the solution. The beauty of the federated system, it's the way the internet was intended to function, oh man I love talking about it. But, nobody unaware of Signal will ever run their own Matrix server of course.

Signal is simple, recognizable, very much like WA. A WA outage is indeed a good point in time to have a lighthearted conversation about messengers and platforms. A lot of people still don't realize Meta owns WA.

What I like is that the normy friends I have on Signal are now telling all their normy friends/colleagues about Signal and how we are still apping away while WA is down.


Call me old fashioned. I email.


Why not IRC? That's old fashion, and amazing :)


Signal is just another centralised architecture.

You should recruit people to Matrix , which supports Federation.


Yep. I host my own server and have it federated. That part was simple.

It was a struggle to get my core group of friends to sign up, and I just sort of haven't bothered trying to get anyone outside that group to try it. They are not technical people in the hn sense but are vastly more competent than the average joe.

It's the normies at large who dictate comms platforms really. Things like matching emoji support and easy + performant video and picture sharing are absolutely crucial.


The EU digital services act will force the big messengers to provide interoperability.


I don't see this happening. Got a link I can learn more about it from?


Join us for AI Startup School this June 16-17 in San Francisco!

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

Search: