It is actually pretty easy to keep alive URLs if there is a will to do so. Just have tests in place so you can't mess them up accidentally. And convert things that are not changing anymore to static html.
One of my first web projects involved a music forum and I have been keeping the URLs alive for 24 years now:
Flattening old things down to plain HTML is underrated and important here. Old apps don’t need to be updated or upgraded or maintained or even deployed. It’s way simpler to argue to keep old stuff around if it’s essentially a domain pointing to a static bucket.
It's not easy. Part of the instructions would need to contain things like "how to convince your boss to not just delete the old thing" and "what to do when the startup you worked for went bankrupt" among others. Most URLs don't disappear by mistake.
Well, and pay registrar and hosting fees for 100 years, and then just hope the hosting company doesn't fold in the interim after your passing. The problem isn't "can you make sure it stays up while you're alive", the problem is, after you're gone, you're no longer there to keep the flame alive and there are no guarantees anyone else picks up the torch.
Does this suggest a subscription(ish) business where you pay the hosting company for N years and then they ensure your website stays up and unchanged for N years after you die? Quality control issue?
Given how many things want me to pay a subscription while I'm alive, surely somebody will start charging me when I'm dead. It is the logical conclusion of SaaS. I wanted to create an acronym for this, but death as a service didn't seem like the best choice.
> It is actually pretty easy to keep alive URLs if there is a will to do so. Just have tests in place
"Just".
Maintaining and keeping these tests (and the oriinal link) running is also probably easy. "Just" keep maintaining them over changes in technology, hosting providers, DNS registrars...
I think that recent LLMs do pass the Turing test, but I don't think it is a good test for intelligence, for two reasons: emergent complexity and pareidolia.
To determine whether a machine is intelligent, I think it's necessary to understand how it works internally and not just take a black-box approach. This was already understood when Alan Turing wrote his 1950 paper. Elmer and Elsie, William Grey Walter's tortoise robots built during the preceding two years, appeared to behave as if they had some degree of self-awareness. But their "brains" consisted of only two thermionic valves, so self-awareness can be ruled out.
I don't know whether LLMs pass the test as Turing originally worded it, but I always interpreted the success condition to be "is indistinguishable from a human" rather than "can trick someone into thinking it's human". Anthropomorphization is a human weakness, and exploiting that doesn't make something intelligent.
I'd only consider an AI to pass the Turing Test if an expert in how the AI works who prepares in advance for the test is unable to tell it apart from a human. The tester should even be allowed to use automated tools as part of the testing process.
If something is indistinguishable from a human in its outward behavior, then I'd be forced to admit that it's as intelligent/conscious/sentient/whatever as a human, regardless of how it works internally. It's not like I make my friends take an MRI before I decide whether they're intelligent.
Most of the LLM tells we usually recognise are due to their heavy attunement to role playing a "helpful assistant" with a knack for correct typography.
If you used a smaller corpus, followed a different kind of fine tuning routine with rather normal conversation patterns, and wrote a different base prompt, I don’t see why it shouldn’t be possible to come up with a very particular personality, so to speak.
> You'd think that LLMs fulfill this, but I do wonder if a clever human could still discern between them given their particularities.
Yes, especially if you use the original adversarial setup with three simultaneous participants (judge, two candidates) as described in Turing's original paper.
However, I'm not quite sure how long even that one will last.
Wayback shows #6 as "By 2010, more than 50 percent of books sold worldwide will be printed on demand at the point of sale in the form of library-quality paperbacks."
It was removed early 2011. Is this what happens when they can't decide on who won?
>You'd think that LLMs fulfill this, but I do wonder if a clever human could still discern between them given their particularities.
While LLMs are very capable of appearing human in general conversation, the test is that they are trying to convince a human that they are human, which is not something that is in their training data and I suspect they would go about things in very odd ways from an actual person, especially when challenged by an actual person. It is still very easy to back LLMs into a corner where they fall into very unnatural patterns and in my experience challenging them on their "humanness" (or anything) backs them into that corner.
LLMs also still can't use context or subtext in natural ways and even people who are terrible with purposefully exploiting context and subtext, generally have a natural ability that they developed to use and parse them. In a short conversation that is not adversarial, this is not an issue but falls apart as the length of the conversation grows or they start getting challenged.
LLMs just don't play with ideas and conversation like we do, they don't riff on it and go with the flow without care, picking up on subtexts and contexts and manipulating them the way people constantly do without effort until they find themselves on the edge of a cliff looking into the void and debating if they should jump. If we are put into the situation of proving we are human (in an academic test sort of way) we will play with it and have fun with it, and we will find ourselves on that cliff looking into the void, debating if we should expose ourselves and admit things we normally wouldn't. We pick up on this when people do it, even over the internet, their way of stating things changes and their subtext falls away or drastically changes; I have never seen anything approaching this in LLMs.
I don't think any frontier lab tried to make an LLM optimized for the literal Turing test where the LLM is made to argue that it's human. This is not in their interest. They make their LLMs insist they aren't human, have no feelings and sentience etc etc.
So we don't actually know how good LLMs can be in this.
> LLMs just don't play with ideas and conversation like we do, they don't riff on it and go with the flow without care, picking up on subtexts and contexts and manipulating them the way people constantly do without effort until they find themselves on the edge of a cliff looking into the void and debating if they should jump.
> You'd think that LLMs fulfill this, but I do wonder if a clever human could still discern between them given their particularities.
I think the clear winning strategy is to find some of the common sense reasoning fails that are still quite common, like the "should I walk or drive to the car wash" one from a few months ago.
These are usually from a missing context (i.e. specifying where the car is right now). Tokenizer issues (how many Xs are in Y) are a better giveaway, but models can solve them using tools nowadays. I think the best way to uncover an LLM is to reveal its superhuman aspects. Ask it for an A* algorithm for your problem. If it can just type down a working algorithm in a few seconds, that is definitely not a human. Although with some careful system prompt design, you could probably mask that aspect too.
In a true competitive Turing test, these reasoning pitfalls could be included in the system prompt, there are not that many of them. The tester would need to find a novel reasoning pitfall, but those aren't uncommon among humans either, so that would not make a very useful differentiator.
Another approach could be to probe the helpfulness of an AI. A human would likely not comply with a request to write a sonnet about tangerines.
The Turing test has been passed. The Turing test had already been passed in 2002. The Turing test isn't about some machine fooling all humans all the time. It's about fooling some humans some of the time. That's an extremely low bar. It has been cleared. Let's put this to rest.
The bar you mentioned was reached in the 1960s with ELIZA. I have no idea why the obsession with the test has persisted into modern generations. It was some guy's thought experiment from 1950. It's not scientific in any fashion. Passing it doesn't prove anything. It's like some kind of authority worship game where people invoke the name of Turing as though doing so bolsters the strength of their arguments in any way whatsoever.
> You'd think that LLMs fulfill this, but I do wonder if a clever human could still discern between them given their particularities.
You don't even have to be that clever. We've all tried using it for work. They fail.
In the first place, the Turing test was always a loosely defined thought experiment. I suppose it's still serving that purpose, but at least half of HN takes it way too seriously. It's absolutely not proof of <engagementbait> endorsed by Alan Turing himself.
I think if you trained an LLM specifically for the purpose of passing the Turing test (instead of being helpful, harmful, and so on), its likely it would pass it.
You would have to train it/finetune it on a couple hundred of 'humans chatting in the context of a Turing test'.
I find it more accurate to refer to "A Turing Test" as opposed to "The Turing Test" for that reason. It's just one loosely-defined test; necessary but not sufficient.
It won't fool all of the people all of the time. But it succeeds often enough to pass the loose definition. It does many things that we called "AI complete" for decades.
It is also clear that is is also not really aware, either. I think Turing would be tickled that we find ourselves in an intermediate state that he would not have imagined.
I would not call Turing's experiment ill-specified, on the contrary among other factors the paper the game is introduced in earnestly calls for a telepathy-proof room to ensure no side channel leakage. The 1950s was an interesting decade.
In my mind, the part that is most likely to fail is "return an HTML document that still contains the following text." Nowadays, more and more "HTML documents" do not contain any content at all, just some JavaScript that is responsible for then fetching and displaying the content.
If the text was returned using a staged process, whereby the server first returns an instruction (JavaScript), and the web browser then executes that instruction, which is what transfers the text, then that doesn't actually change anything. Those are just technical details of the transfer. You are viewing an HTML document on-screen; it makes no difference to this bet if it was returned to you by a different mechanism than originally conceived, since the bet doesn't specify any transfer mechanism, just that the URL was entered into a browser.
I think the parent is referring to the possibility that one day, browsers might put up a blanket security warning for HTTP URLs instead of following redirects. Or they might try to be a little too clever with HTTPS upgrades.
In particular, the automatic upgrade feature in modern browsers is based on several heuristics rather than explicit configuration like HSTS, so there's always a bit of room for breakage there. For example, they don't even check if the server returns a 301 redirect, which can be problematic if the server wants to redirect to HTTPS on a different host/port or make some changes to the path.
At a certain bet size it would almost be self-fulfilling, heh, as one side would be very incentivized to keep at least that url working (buying the domain + hosting it if the page were to die)
I was expecting some kind of system that would follow the pages automatically without you having to do extra work when changing the URL (“continue to work even when I change them and forget them”) but it looks like you still add the redirects manually?
That’s more than most will do, sure, but it doesn’t seem like it saves you from a broken link if you forget to add a specific redirect or miss it during a mass format change. Am I missing something?
I really liked disqus back in the day. When I was in my kinda cringey singularity phase, Kurzweil and all that. Tracking comments across similar blogs with the same people was handy.
I know nothing of Safecloud, but freenet failed for a very valid reasons:
1. Data retention isn't infinite. If the source disappears from the network, peers will prune it shortly. This is exponentially true in our multimedia-heavy world.
2. Despite 1, if you actually want to delete something you published, you can't. Absolutists will see this as a good thing. Normal people will see this as a problem.
3. Peer-to-peer doesn't work in a mobile-first world, and not enough people will host nodes as a hobby. If all website publishers/hosters have to host a node, then we're just back to the current model but with extra steps.
It is actually pretty easy to keep alive URLs if there is a will to do so. Just have tests in place so you can't mess them up accidentally. And convert things that are not changing anymore to static html.
One of my first web projects involved a music forum and I have been keeping the URLs alive for 24 years now:
https://www.gnoosic.com/discussion/
I feel like I owe it to the people who participated to keep it online forever. Also as a document of history.
Cool URIs don't change, as we should all remember:
https://www.w3.org/Provider/Style/URI
Flattening old things down to plain HTML is underrated and important here. Old apps don’t need to be updated or upgraded or maintained or even deployed. It’s way simpler to argue to keep old stuff around if it’s essentially a domain pointing to a static bucket.
I did that to palmpowerups.com. The original CMS is long gone.
It's not easy. Part of the instructions would need to contain things like "how to convince your boss to not just delete the old thing" and "what to do when the startup you worked for went bankrupt" among others. Most URLs don't disappear by mistake.
Well, and pay registrar and hosting fees for 100 years, and then just hope the hosting company doesn't fold in the interim after your passing. The problem isn't "can you make sure it stays up while you're alive", the problem is, after you're gone, you're no longer there to keep the flame alive and there are no guarantees anyone else picks up the torch.
Does this suggest a subscription(ish) business where you pay the hosting company for N years and then they ensure your website stays up and unchanged for N years after you die? Quality control issue?
Given how many things want me to pay a subscription while I'm alive, surely somebody will start charging me when I'm dead. It is the logical conclusion of SaaS. I wanted to create an acronym for this, but death as a service didn't seem like the best choice.
3 replies →
*a will and the ability for that will to access what it needs
> I feel like I owe it to the people who participated to keep it online forever. Also as a document of history.
Oh! So its your website. Cool.
Did the money change hands, for this bet, in 2022? How do you ensure it. Or is it honor based?
> It is actually pretty easy to keep alive URLs if there is a will to do so. Just have tests in place
"Just".
Maintaining and keeping these tests (and the oriinal link) running is also probably easy. "Just" keep maintaining them over changes in technology, hosting providers, DNS registrars...
It's worth noting that some prediction URLs, for instance: https://longbets.org/6/ are offline.
The original long bet was whether or not computers could pass the turing test by 2029: https://longbets.org/1/
You'd think that LLMs fulfill this, but I do wonder if a clever human could still discern between them given their particularities.
https://longbets.org/9/ is another interesting one depending on whether you think covid leaked from a lab or not.
I think that recent LLMs do pass the Turing test, but I don't think it is a good test for intelligence, for two reasons: emergent complexity and pareidolia.
Arguably, ELIZA passed the Turing test ca. 1967: https://archive.computerhistory.org/resources/access/text/20... and not that long ago, ELIZA was still more convincing: https://arstechnica.com/information-technology/2023/12/real-...
The Turing Test can be gamed. This was done in some Loebner Prize entries: see https://www.behavior.org/resources/325.pdf and https://en.wikipedia.org/wiki/Eugene_Goostman
To determine whether a machine is intelligent, I think it's necessary to understand how it works internally and not just take a black-box approach. This was already understood when Alan Turing wrote his 1950 paper. Elmer and Elsie, William Grey Walter's tortoise robots built during the preceding two years, appeared to behave as if they had some degree of self-awareness. But their "brains" consisted of only two thermionic valves, so self-awareness can be ruled out.
I don't know whether LLMs pass the test as Turing originally worded it, but I always interpreted the success condition to be "is indistinguishable from a human" rather than "can trick someone into thinking it's human". Anthropomorphization is a human weakness, and exploiting that doesn't make something intelligent.
I'd only consider an AI to pass the Turing Test if an expert in how the AI works who prepares in advance for the test is unable to tell it apart from a human. The tester should even be allowed to use automated tools as part of the testing process.
If something is indistinguishable from a human in its outward behavior, then I'd be forced to admit that it's as intelligent/conscious/sentient/whatever as a human, regardless of how it works internally. It's not like I make my friends take an MRI before I decide whether they're intelligent.
2 replies →
There was a paper a while ago, where they found a strategy for GPT beating humans on the Turing test 2/3rds of the time.
They just prompted it to talk the way a 13 year old girl texts.
(Also they prompted it to pretend not to know anything about anything.)
Most of the LLM tells we usually recognise are due to their heavy attunement to role playing a "helpful assistant" with a knack for correct typography.
If you used a smaller corpus, followed a different kind of fine tuning routine with rather normal conversation patterns, and wrote a different base prompt, I don’t see why it shouldn’t be possible to come up with a very particular personality, so to speak.
> where they found a strategy for GPT beating humans on the Turing test 2/3rds of the time.
IIRC, PARRY https://simulation-and-society.org/parry/ got there in 1972 by pretending to be a paranoid schizophrenic person
2 replies →
I feel like pretending to be a child is cheating somewhat. Might as well print random key presses and claim it's a toddler.
1 reply →
> You'd think that LLMs fulfill this, but I do wonder if a clever human could still discern between them given their particularities.
Yes, especially if you use the original adversarial setup with three simultaneous participants (judge, two candidates) as described in Turing's original paper.
However, I'm not quite sure how long even that one will last.
>https://longbets.org/6/
Anyone know why this (and many other bets) have been removed?
https://web.archive.org/web/20080803213625/https://longbets....
>By 2010, more than 50 percent of books sold worldwide will be printed on demand at the point of sale in the form of library-quality paperbacks.
Not exactly a hot-button issue.
Wayback shows #6 as "By 2010, more than 50 percent of books sold worldwide will be printed on demand at the point of sale in the form of library-quality paperbacks."
It was removed early 2011. Is this what happens when they can't decide on who won?
>You'd think that LLMs fulfill this, but I do wonder if a clever human could still discern between them given their particularities.
While LLMs are very capable of appearing human in general conversation, the test is that they are trying to convince a human that they are human, which is not something that is in their training data and I suspect they would go about things in very odd ways from an actual person, especially when challenged by an actual person. It is still very easy to back LLMs into a corner where they fall into very unnatural patterns and in my experience challenging them on their "humanness" (or anything) backs them into that corner.
LLMs also still can't use context or subtext in natural ways and even people who are terrible with purposefully exploiting context and subtext, generally have a natural ability that they developed to use and parse them. In a short conversation that is not adversarial, this is not an issue but falls apart as the length of the conversation grows or they start getting challenged.
LLMs just don't play with ideas and conversation like we do, they don't riff on it and go with the flow without care, picking up on subtexts and contexts and manipulating them the way people constantly do without effort until they find themselves on the edge of a cliff looking into the void and debating if they should jump. If we are put into the situation of proving we are human (in an academic test sort of way) we will play with it and have fun with it, and we will find ourselves on that cliff looking into the void, debating if we should expose ourselves and admit things we normally wouldn't. We pick up on this when people do it, even over the internet, their way of stating things changes and their subtext falls away or drastically changes; I have never seen anything approaching this in LLMs.
I don't think any frontier lab tried to make an LLM optimized for the literal Turing test where the LLM is made to argue that it's human. This is not in their interest. They make their LLMs insist they aren't human, have no feelings and sentience etc etc.
So we don't actually know how good LLMs can be in this.
2 replies →
> LLMs just don't play with ideas and conversation like we do, they don't riff on it and go with the flow without care, picking up on subtexts and contexts and manipulating them the way people constantly do without effort until they find themselves on the edge of a cliff looking into the void and debating if they should jump.
Sounds like Neuro-sama to me.
> You'd think that LLMs fulfill this, but I do wonder if a clever human could still discern between them given their particularities.
I think the clear winning strategy is to find some of the common sense reasoning fails that are still quite common, like the "should I walk or drive to the car wash" one from a few months ago.
These are usually from a missing context (i.e. specifying where the car is right now). Tokenizer issues (how many Xs are in Y) are a better giveaway, but models can solve them using tools nowadays. I think the best way to uncover an LLM is to reveal its superhuman aspects. Ask it for an A* algorithm for your problem. If it can just type down a working algorithm in a few seconds, that is definitely not a human. Although with some careful system prompt design, you could probably mask that aspect too.
4 replies →
In a true competitive Turing test, these reasoning pitfalls could be included in the system prompt, there are not that many of them. The tester would need to find a novel reasoning pitfall, but those aren't uncommon among humans either, so that would not make a very useful differentiator.
Another approach could be to probe the helpfulness of an AI. A human would likely not comply with a request to write a sonnet about tangerines.
1 reply →
By 2029 things will be different. Humans will have learned to talk like LLMs.
The Turing test has been passed. The Turing test had already been passed in 2002. The Turing test isn't about some machine fooling all humans all the time. It's about fooling some humans some of the time. That's an extremely low bar. It has been cleared. Let's put this to rest.
The bar you mentioned was reached in the 1960s with ELIZA. I have no idea why the obsession with the test has persisted into modern generations. It was some guy's thought experiment from 1950. It's not scientific in any fashion. Passing it doesn't prove anything. It's like some kind of authority worship game where people invoke the name of Turing as though doing so bolsters the strength of their arguments in any way whatsoever.
> You'd think that LLMs fulfill this, but I do wonder if a clever human could still discern between them given their particularities.
You don't even have to be that clever. We've all tried using it for work. They fail.
In the first place, the Turing test was always a loosely defined thought experiment. I suppose it's still serving that purpose, but at least half of HN takes it way too seriously. It's absolutely not proof of <engagementbait> endorsed by Alan Turing himself.
I think if you trained an LLM specifically for the purpose of passing the Turing test (instead of being helpful, harmful, and so on), its likely it would pass it.
You would have to train it/finetune it on a couple hundred of 'humans chatting in the context of a Turing test'.
4 replies →
Mannerisms aside, they are easy to spot from having inhuman amounts of trivia knowledge
2 replies →
I find it more accurate to refer to "A Turing Test" as opposed to "The Turing Test" for that reason. It's just one loosely-defined test; necessary but not sufficient.
It won't fool all of the people all of the time. But it succeeds often enough to pass the loose definition. It does many things that we called "AI complete" for decades.
It is also clear that is is also not really aware, either. I think Turing would be tickled that we find ourselves in an intermediate state that he would not have imagined.
I would not call Turing's experiment ill-specified, on the contrary among other factors the paper the game is introduced in earnestly calls for a telepathy-proof room to ensure no side channel leakage. The 1950s was an interesting decade.
> entering the characters http://www.longbets.org/601 into the address bar of a web browser or command line tool (like curl)
The part of this that's most likely to fail at some point is the "http://".
In my mind, the part that is most likely to fail is "return an HTML document that still contains the following text." Nowadays, more and more "HTML documents" do not contain any content at all, just some JavaScript that is responsible for then fetching and displaying the content.
If the text was returned using a staged process, whereby the server first returns an instruction (JavaScript), and the web browser then executes that instruction, which is what transfers the text, then that doesn't actually change anything. Those are just technical details of the transfer. You are viewing an HTML document on-screen; it makes no difference to this bet if it was returned to you by a different mechanism than originally conceived, since the bet doesn't specify any transfer mechanism, just that the URL was entered into a browser.
5 replies →
You two have just spelled out a very specific bet. Care to put a wager on it? :)
I used to be someone, man, I was a dreamer, a programmer, a man!
Now I'm a human slave copying text from web browsers to chat windows to get around JS rendering bullshit.
4 replies →
The detailed terms of the bet include this:
> A 301 redirect from www.longbets.org/601 to a different URL containing that text would also fulfill those conditions.
I assume this would cover the HTTP->HTTPS redirect scenario.
I think the parent is referring to the possibility that one day, browsers might put up a blanket security warning for HTTP URLs instead of following redirects. Or they might try to be a little too clever with HTTPS upgrades.
In particular, the automatic upgrade feature in modern browsers is based on several heuristics rather than explicit configuration like HSTS, so there's always a bit of room for breakage there. For example, they don't even check if the server returns a 301 redirect, which can be problematic if the server wants to redirect to HTTPS on a different host/port or make some changes to the path.
2 replies →
Yeah that was a clever bit of foresight there.
At a certain bet size it would almost be self-fulfilling, heh, as one side would be very incentivized to keep at least that url working (buying the domain + hosting it if the page were to die)
Someone should do a long bet on the continuous existence of longbets.com
One thing I've done is include my redirect rules with the web application. So my links continue to work even when I change them and forget them.
https://news.ycombinator.com/item?id=46139686
I was expecting some kind of system that would follow the pages automatically without you having to do extra work when changing the URL (“continue to work even when I change them and forget them”) but it looks like you still add the redirects manually?
That’s more than most will do, sure, but it doesn’t seem like it saves you from a broken link if you forget to add a specific redirect or miss it during a mass format change. Am I missing something?
Amazing that Disqus comments are still live for this page. More impressive even than the website itself being live.
I really liked disqus back in the day. When I was in my kinda cringey singularity phase, Kurzweil and all that. Tracking comments across similar blogs with the same people was handy.
I'm grateful the comments are still there, since they reminded me to get back in touch with my old friend Craig Hubley!
Some nice posts about his prediction on Jeremy Keith's blog:
https://adactio.com/journal/tags/longbets
Wow, that could be a post of it’s own
Anyway to find all bets expiring this year? it will be interesting to see. Could not find a way to search for it there.
I'm not sure how it played out in 2011, but as of now it looks like the win condition hinges on the redirect.
For the heck of it I confirmed "http://www.longbets.org/601" does in fact result in a 301 redirect to "https://longbets.org/601/". I suspected they might have used a 302.
Ugh, I really hate SSL. This is a simple text page, why couldn't it live where it was?
Yes, I'm an old fart, yelling at the internet clouds. You'll get here some day too, if you're lucky.
I am quite impressed that the bet included above redirection as part of the terms.
This is basically a bet on if you think longbets is going to be around.
I like the comment on the site about blockchains replacing HTML links (from 4 years ago).
It's all a matter of whether someone is willing to pay for the hosting.
Which is not, in my opinion, how things should be hosted on the Web.
Freenet had a different approach, even back in the early days (two decades ago). I interviewed Ian Clarke about it https://www.youtube.com/watch?v=JWrRqUkJpMQ
I wrote about it for years: https://community.intercoin.app/t/who-pays-for-storage-nfts-...
And recently I built Safecloud to be different: https://safebots.github.io/Safecloud/
On HN, a TLDR would be helpful, because just clicking your links would potentially mean falling for shilling of ones own side-hustle.
I bet there's complexity to the topic, but I am sure you could put it into a few words instead of just leaving a hook for others to click on.
I know nothing of Safecloud, but freenet failed for a very valid reasons:
1. Data retention isn't infinite. If the source disappears from the network, peers will prune it shortly. This is exponentially true in our multimedia-heavy world.
2. Despite 1, if you actually want to delete something you published, you can't. Absolutists will see this as a good thing. Normal people will see this as a problem.
3. Peer-to-peer doesn't work in a mobile-first world, and not enough people will host nodes as a hobby. If all website publishers/hosters have to host a node, then we're just back to the current model but with extra steps.
3 replies →
[flagged]
[dead]
[dead]