Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

No clicks, no content: The unsustainable future of AI search

bradt.ca

Post reply on HN
Print view View original

171–180 of 190 posts

Re: No clicks, no content: The unsustainable future of AI search

#171

Post by zenolijo »

amarant wrote: post #105
fabian2k wrote: post #64

Earlier quoted context omitted.

But AI is also going to kill some of your positive examples. Stack Overflow for example is in a steep decline, only a small fraction of questions are posted today compared to the peak. And the effects are more than financial, so even non-profit examples like forums would be hit. If new people don't discover your site with useful user-created content, they won't contribute to it. You're also cutting off the pipeline f…

Surely part of that is because most tech related questions have already been asked and answered on SO. I'd say a decline in new question is stack overflow working as designed. A large part of what makes SO so good is the searchability of old questions. There will always be new questions to ask, as new technologies confuse Devs in new ways, but to expect new questions to be asked at the same rate as peak is to misunde…

With one quite significant issue IMO

If there's an old question the most upvoted answer will be at the top. Better solutions are often available if the previous answer was 10 years ago, but they will be buried.

Solution is obviously to scroll down as well as read the comments, but that can be time consuming.

Re: No clicks, no content: The unsustainable future of AI search

#172

Post by BoredPositron »

verdverm wrote: post #74
fabian2k wrote: post #64

Earlier quoted context omitted.

But AI is also going to kill some of your positive examples. Stack Overflow for example is in a steep decline, only a small fraction of questions are posted today compared to the peak. And the effects are more than financial, so even non-profit examples like forums would be hit. If new people don't discover your site with useful user-created content, they won't contribute to it. You're also cutting off the pipeline f…

Stuck overflow may not be the greatest example. I have switch to using GitHub discussions and Discord on the "where to get help for my projects" side of things. I ignore SO when it comes to support. Lots of other open source projects doing similar. This trend was happening before LLMs entered the arena.

I am with you on this the amount of answers I got from people that actually have knowledge about the libraries I am working with was always overshadowed by some semantic questioning. Do really need to? This tool might be better etc.

Re: No clicks, no content: The unsustainable future of AI search

#173

Post by BoredPositron »

natebc wrote: post #85
roblh wrote: post #81

Earlier quoted context omitted.

Discord is just absolutely worthless for this. Any question that gets asked gets buried in days if not hours. It pretty much guarantees the same basic garbage gets repeated over and over and over forever. Basically the exact opposite of stack overflow.

Inevitably too you'll get someone scolding you to "check the pins" which you then do and get introduced to that hellish nightmare. Discord is great for chatting with your friends, gaming, etc. but man it's a horrible knowledge repository.

That's why I still prefer IRC without retention there is more freedom for discussion.

Re: No clicks, no content: The unsustainable future of AI search

#174

Post by rainsford »

kleiba wrote: post #13

The argument seems flawed to me: by "killing the web", they refer to the example of a company adding SEO'd information to their website to lure in traffic from web searches. However, me personally, I don't want to be lured into some web store when I'm looking for some vaguely related information. Luckily, there's tons of information on the web provided not by commercial entities but by volunteers: wikipedia, forum us…

I suppose wanting to kill the commercial web is a valid position, although it feels more like grumpy old man yells at kids to get off his lawn than a considered analysis of relative value and impact, but even then I think you're underestimating the impact the AI problem will have on the non-commercial web as well.

Lots of people might be willing to run websites for fun or personal satisfaction or whatever, but how many people will continue to be willing to do so when they don't actually get to present the content to visitors and it's instead just regurgitated by AI? Half the fun of hosting your own website is personalizing it and choosing how to share the content. Even people blogging for fun put a lot of thought into their posts on how to phrase an argument or tell a story. But what's the point when nobody will ever see your actual post, just your thoughts rearranged and presented by AI? Maybe some people only care about the information being out there in any form, but I'd be willing to bet that's yet a smaller subset of even the people who would contribute in a return to geocities version of the web.

Re: No clicks, no content: The unsustainable future of AI search

#175

Post by balder1991 »

maltelandwehr wrote: post #160
balder1991 wrote: post #49

Earlier quoted context omitted.

What I notice here in Brazil is that most local news channels get the bulk of their money from TV ads. They all have a badly done website-blog with news that are very superficial (like 2 paragraphs) just to fill them with ads up and down and try to get something from it. The big channels nowadays usually have 2 websites: one that is free and full of ads and pop-ups with very superficial news (seemingly written by int…

How can we adapt that to LLMs? Do LLM providers pay for access to these articles? Do I as a user have to do a micro transaction whenever an LLM generates an answer on one of those paywalled articles? Because as a user, I do not wish to read the quality journalist analysis, I wish for it to be part of the LLM answer that is tailored towards me.

I think the micro transactions thing would be ideal, and should be something between the LLM providers and the websites, the users should pay solely for the subscription of LLMs being able to search.

But this is a huge simplification of course, and another thousand problems arise from this model. So I have no idea what’s the “good enough” solution we’ll head towards, or whether the web will change completely from this.

Re: No clicks, no content: The unsustainable future of AI search

#176

Post by visarga »

_DeadFred_ wrote: post #157
visarga wrote: post #38

The content is now created in private chats with AI, probably a trillion tokens per day flow between humans and LLMs. When new AI models are made they will incorporate some experience from past usage. This dataset might be the most valuable source for training AI and solving our tasks. So in case humans decide to abandon publishing, there is a new experience flywheel spinning up.

So they will learn on the hallucinations they told to people? Sounds like a sound system.

What happens when AI helps you on a task? You usually use its outputs to do something, and if that works, you come back for further assistance. If it doesn't work, you come back to correct the model. Either way a signal from the real world gets captured in the chat logs.

When AI provides a response it is possible to judge that response in hindsight. You look at the next 20 messages or sessions from next days and judge based on what followed. The chat logs provide a way to do long range credit assignment.

Re: No clicks, no content: The unsustainable future of AI search

#177

Post by visarga »

nicbou wrote: post #63
visarga wrote: post #38

The content is now created in private chats with AI, probably a trillion tokens per day flow between humans and LLMs. When new AI models are made they will incorporate some experience from past usage. This dataset might be the most valuable source for training AI and solving our tasks. So in case humans decide to abandon publishing, there is a new experience flywheel spinning up.

Where will the experience come from?

The AI model chats with the human, and the human uses AI generations in the real world. The real world provides feedback through the human back to AI.

Re: No clicks, no content: The unsustainable future of AI search

#178

Post by visarga »

8organicbits wrote: post #50
visarga wrote: post #38

The content is now created in private chats with AI, probably a trillion tokens per day flow between humans and LLMs. When new AI models are made they will incorporate some experience from past usage. This dataset might be the most valuable source for training AI and solving our tasks. So in case humans decide to abandon publishing, there is a new experience flywheel spinning up.

I assume that AI chat is a lot of questions being asked by people unfamiliar with the subject matter. Would that training dataset have any information from experts?

Question answering and learning are just a corner of LLM usage, but they have learning signals for the AI. Say a user asks about Pythagoras, the LLM provides an explanation, the user doesn't get it. The LLM tries again.

Repeat this loop a million times with diverse students and you get a distribution of what kind of explanations work. The model gets better at explaining through its own experience.

Re: No clicks, no content: The unsustainable future of AI search

#179

Post by 8organicbits »

visarga wrote: post #178
8organicbits wrote: post #50

Earlier quoted context omitted.

I assume that AI chat is a lot of questions being asked by people unfamiliar with the subject matter. Would that training dataset have any information from experts?

Question answering and learning are just a corner of LLM usage, but they have learning signals for the AI. Say a user asks about Pythagoras, the LLM provides an explanation, the user doesn't get it. The LLM tries again. Repeat this loop a million times with diverse students and you get a distribution of what kind of explanations work. The model gets better at explaining through its own experience.

Sounds like you'd end up with pop science. The loop stops when the explanation is satisfying, not when it's correct. Vibe science isn't based in reality.

Re: No clicks, no content: The unsustainable future of AI search

#180

Post by carlosjobim »

Waterluvian wrote: post #60

If we can streamline the generalized information seeking process, that part of the web can dry up and disappear. And then we’d be left with more of the early era web, where you’re visiting websites not because you have a specific question to answer, but because you’re engaging in a social or interactive or otherwise deeper activity. When it comes to “I have a specific question I need answering and then I’m done” the…

The early era of the web was an exclusive club for well-paid computer engineers and other computer specialists with a lot of spare time. Which was reflected in information being heavily focused on certain topics and perspectives heavily tilted to certain types of people with a narrow band of ideologies and beliefs. Just like Hacker News is today.

I prefer a million times today's web, which serves everybody and where I can find all kind of thoughts and ideas, without restriction. You just need to make an effort to find it. I prefer a million times a well stocked supermarket with all the ingredients I need to make anything I want, rather than a restaurant which serves only one meal made perfectly.

Post reply on HN