I was still digesting the throat-clearing intro when my eye noticed "quietly" and "honest" and I decided I'd not bother. But I scrolled through out of politeness, just in case the Basilisk will be very thorough.
> I want to be clear that I'm not against these tools generally. I use them myself
No essay is made worse by being stripped of filler. If the reader has got this far, this sentence is entirely redundant, I think.
I'm against them, vehemently, but I'm also not going to fight a tidal wave. I'm not going to be unemployed and unemployable by refusing to use them. Principles are expensive and I certainly cannot realistically afford to hold them tightly.
Being forced to use them to continue to have an income is causing me a cognitive dissonance and a level of stress that is surely going to kill me though. I strongly resent the fact that I've been backed into this corner where my choice is to swallow my principles or throw away a career I've worked half my life on
Even if it's not inevitable, if your employer tells you "use AI or get fired", well, take your pick. A lot of the resentment I see is a result of those sorts of mandates.
That’s why I don’t entertain their fatalism. We’re all rats on a sinking ship powerless to do anything outside of our existing schedules. Indistinguishable outcome for the majority of people compared to the AI euphorics.
I do worry about Google AI overview's long-term impact, but man, it has been such a great experience of late. It surely does feel like the old google but better.
There was a point, perhaps about five years ago, where using Google had become a genuinely frustrating hurdle I needed to overcome to do the actual work. Now the frustration is completely gone.
Just this morning I had a worrying result when I googled "UV index <location>" and it confidently told me the current UV index was zero.
It was not, and at least because it was obviously wrong I clicked through to a page with the real UV index.
But had it said "The UV index is 1.7" I might have believed it, but it'd still have been wrong, potentially dangerously so.
Now to be clear, live data is precisely the sort of thing I expect it to get wrong, and I never intended to trigger let alone read an "AI summary", but it's still worrying for those less inclined to dismiss it.
They used to handle real data extremely well by using those little search widgets they would recognize what you’re asking and just show weather etc. the AI answers is just such a massive regression.
It would be obviously wrong if location=out standing in a middle of a field in direct sunlight. It wouldn't be obviously wrong if location=indoors.
Proving that search hasn't stopped making us think at all. Just like we've had to do with humans since the dawn of time, we continue to have to think hard about how our intent will be interpreted.
It's almost like what we really need is someone to simply index the sites that have the actual primary sources so we can search them and find them quickly! /s
Interesting given this is the polar opposite of my experience. I find Googles AI overview to give wrong answers constantly. Occasionally, even for questions I believe Google circa 2018 was able to answer confidently.
Ironically, Googles AI overview often links to references that state the polar opposite of what the AI overview claims.
I've seen it answering based on a single Reddit, Stackexchange, or even HN source. Sometimes either incorrectly, or taking some side of a debated technical issue and presenting it like a fact.
No way, Google AI search always says you need to see a medical professional or if you say your cat puked it’s like “Cancer” it’s basically as bad as early Google always wrong about everything and over-worrying and over-assuming.
Half the time it just describes some unrelated shit. Maybe it’s interesting but totally sidetracks your search.
IMO it’s like Google 15 years ago where it was causing more panic and disinformation but people knew it back then - we laughed at Google’s persistent and incorrect diagnoses.
Now though, with boomers and young Gen Y and older Gen Z - maybe the dumbest yet most confident people alive it’s causing a lot of harm. Everyone thinks they know everything and because they inherit from their parents and are basically rich society rewards them for being retarded
I have started filtering for pre-2020 items when I am searching for articles.
Obviously not a silver bullet - for example, not much help if you're looking for modern tooling-related stuff - but I've found it really useful as a way of reading up on fundamentals.
I can't be the only one that scrolls past quickly the whatever Google-AI spurts out. If I wanted to ask an AI - I would. But I've learned that they hallucinate stuff with numbers or anything that's time sensitive/recent (even if that might not be the truth now). So I won't spend my time on verifying their output. I'm after a quick search, not a likely guess.
What's worse, when reading the Google outputs, I have learned to ignore almost any website besides a select few, one of which is Reddit. Which is very weird thinking about it, but it worked in the past.
The last thing I searched Google for was "india gdp growth" and Google gave me an estimate for India's 2026 GDP growth. It took like 2 seconds. It's almost certainly basically correct.
If I had tried to find an authoritative source I might have spent a minute on it. Would it have killed me? No. Would I do that for something more life-threatening? Of course. But flash results are... fine, for 90% of what I search for.
Fundamental learning path does not change.
Whatever searche result force feeds you it's ok.
So long as you have a critical thinking mindset.
People simply read titles, and oh that's what happened.
If you apply the first order thinking, you can spot most of mistakes in docs and results.
But then you gotta move onto the second order thinking to see if the conclusion you came up with will be good for the future.
Say, you searched for how to build a house yourself.
You see an AI result that recommends contractors.
But if you break it down why AI recommended the contractor, you can see harness (assumptions) such ass security (the builder/searcher should not be harmed), convinience (how fast do you need) without regards to your personal needs.
- that's why comapnies want so much of your info, so they can try to guess better output based on your needs.
Now you breakdown using the 1st principle, you find out that you don't want those assumed AI answer, and you ask more direct question and guide it.
But then after applying the 2nd order principle, you might buy into the AI's recommendation that you don't want to spend too much time building one yourself for the safety (you are a pro athelet, who should not do heavy labor for the sake of the next game match).
Anyways, consider multiple thinking models and views.
Has anyone tried making a p2p search engine? Naively it seems like one of the p2p networks from the late 90s or early 2000s should be able to handle it.
Searches propagate between nodes, nodes specialize on specific sections of the index, which can be random + user preference (maximizing redundance and availability for the things people actually care about).
Soulseek already gets like halfway there and it's not even trying to be. How hard could it be?
So my MS research focused on P2P networks, and with the caveat that the area has been out of favor (and hence under-researched) for about 2 decades, I would say it would be extremely hard to build even decent quality search in a purely decentralized manner.
There were essentially two main ways of doing P2P search: flooding and Distributed Hash Tables (DHT). Flooding is exactly how it sounds: you flood a query across the whole network. The most popular applications like Limewire, Kazaa etc were flooding based, and as you can imagine, didn’t scale well. I recall some stories of havoc on networks due to the amount of traffic they generated. However they were simple and allowed somewhat complex queries, because each node basically ran a brute force search on its local index.
On the other hand, DHTs organize nodes in a distributed index structure such that queries can be routed in a deterministic manner, typically in O(logN) hops. DHTs were basically the granddaddies of distributed NoSQL DBs like DynamoDB and Cassandra.
So you can already see the limitations of DHTs: they need keys to be indexed, typically keywords. You had to pick a small handful of keywords for it to be manageable. This inherently limits the complexity and scale of the queries.
Both had an existential problem with spam. There were essentially no defenses against bad actors poisoning the network with spam and malware. It was a huge unsolved problem until Bitcoin popularized proof of work (which had been suggested in various forms before, but Bitcoin was the first practical, robust application.)
And neither could approach the sophistication of Google, which could extract very high quality signals from their centralized troves of data to refine and rank results.
Personally though I think the time is ripe for a renaissance. Spam and bad actors are still existential problems, but with vectors / embeddings the discovery aspect could be addressed. It occurs to me that the most popular large scale vector search mechanism is HNSW networks, which seem pretty amenable to P2P networks…
YaCy[0] is a P2P search engine. There's public instances but you can self-host it as well. It includes a spider to do your own crawling. It's challenging now between CloudFlare and other anti-bot CAPCHAS crawling effectively is really challenging. Bots (AI and otherwise) have ruined the web.
This is a very well trodden argument the poster is largely right about. But like others who have written about it like Nicholas Carr and Anil Dash, I agree with the concern, but it places a lot of agency on the technology and not the users decisions. On one hand, you can argue for a long time Google was the default page for the internet so you had no choice but to use it to search, on the other hand, people need to use their brains more when thinking critically. They need to learn how to use it to assist in thought. Yes, Google search results are ads, but scroll past the first few pages and dig. Also, refine searches and look at other sources by people you trust. Ultimately, this is a real concern, and I'm glad the author is updating thoughts on it.
It is a naive question but I do wonder how hard it is to scrape the web and build web indexes that are shared via bit torrent and self host a tfidf traditional engine using a distributed effort. How big the web really is?
Is there a lot of additional secret sauce that made google work well in its prime?
I don't think it hurts thinking (at least I haven't noticed anything on myself yet), but it hurts the internet a lot, and may even destroy it in parallel to LLM.
Online search became completely unusable, and it started well before chatgpt arrived.
Now when I am looking for information on this or that I'm obliged to append reddit, Wikipedia or YouTube in my search to hope to have something with real content (and I hate watching videos to be kept informed and I feel there is more and more LLM generated content there also).
I even have a subscription to chatgpt now that I use more and more as a replacement for a Google search, or at least as a supplement.
I don't like it very much because in the best case I have only a handful of sources that it selected instead of being able to look by myself at all the possible sources, but it is still better than spending hours finding something that is not SEO-slop.
I'm convinced it will never improve because it helps google and the others push their AI stuff, and they desperately need it to justify the billions poured in it, but also because it gave them a much deeper control on what is provided to the users.
Yeah, I agree with what you are saying, people seem to take the AI response at face value. I answer lies in the users recognizing that HAL and others can, and right now often do, make mistakes.
I‘m outsourcing all tedious searches to ChatGPT now. Anytime I try searching anything non-obvious, it seems to be buried in the Nth page. Both Google and DDG have this problem. Chat spares me the brain power of sifting through the endless slop and grift, and I can easily click through and read the actual content.
I was still digesting the throat-clearing intro when my eye noticed "quietly" and "honest" and I decided I'd not bother. But I scrolled through out of politeness, just in case the Basilisk will be very thorough.
> I want to be clear that I'm not against these tools generally. I use them myself
No essay is made worse by being stripped of filler. If the reader has got this far, this sentence is entirely redundant, I think.
"I am against these tools generally. I use them myself" would be a more interesting take
I am against murdering animals. But I eat animals.
I'm against them, vehemently, but I'm also not going to fight a tidal wave. I'm not going to be unemployed and unemployable by refusing to use them. Principles are expensive and I certainly cannot realistically afford to hold them tightly.
Being forced to use them to continue to have an income is causing me a cognitive dissonance and a level of stress that is surely going to kill me though. I strongly resent the fact that I've been backed into this corner where my choice is to swallow my principles or throw away a career I've worked half my life on
Pro AI: It’s inevitable and I love it
Anti AI: It’s inevitable and I hate it
Oh no. They’re converging. But I guess that just proves the point doesn’t it.
Even if it's not inevitable, if your employer tells you "use AI or get fired", well, take your pick. A lot of the resentment I see is a result of those sorts of mandates.
That’s why I don’t entertain their fatalism. We’re all rats on a sinking ship powerless to do anything outside of our existing schedules. Indistinguishable outcome for the majority of people compared to the AI euphorics.
Yeah, glazed immediately. It's not even like I recognised it as generated, there's just no substance.
I do worry about Google AI overview's long-term impact, but man, it has been such a great experience of late. It surely does feel like the old google but better.
There was a point, perhaps about five years ago, where using Google had become a genuinely frustrating hurdle I needed to overcome to do the actual work. Now the frustration is completely gone.
Just this morning I had a worrying result when I googled "UV index <location>" and it confidently told me the current UV index was zero.
It was not, and at least because it was obviously wrong I clicked through to a page with the real UV index.
But had it said "The UV index is 1.7" I might have believed it, but it'd still have been wrong, potentially dangerously so.
Now to be clear, live data is precisely the sort of thing I expect it to get wrong, and I never intended to trigger let alone read an "AI summary", but it's still worrying for those less inclined to dismiss it.
They used to handle real data extremely well by using those little search widgets they would recognize what you’re asking and just show weather etc. the AI answers is just such a massive regression.
The worst example of this seems to be tide tables, which are extremely date-dependant.
It would be obviously wrong if location=out standing in a middle of a field in direct sunlight. It wouldn't be obviously wrong if location=indoors.
Proving that search hasn't stopped making us think at all. Just like we've had to do with humans since the dawn of time, we continue to have to think hard about how our intent will be interpreted.
It's almost like what we really need is someone to simply index the sites that have the actual primary sources so we can search them and find them quickly! /s
Interesting given this is the polar opposite of my experience. I find Googles AI overview to give wrong answers constantly. Occasionally, even for questions I believe Google circa 2018 was able to answer confidently.
Ironically, Googles AI overview often links to references that state the polar opposite of what the AI overview claims.
I've seen it answering based on a single Reddit, Stackexchange, or even HN source. Sometimes either incorrectly, or taking some side of a debated technical issue and presenting it like a fact.
Each Google "AI" response should state a correctness confidence value. A range 0-40% will probably suffice.
No way, Google AI search always says you need to see a medical professional or if you say your cat puked it’s like “Cancer” it’s basically as bad as early Google always wrong about everything and over-worrying and over-assuming.
Half the time it just describes some unrelated shit. Maybe it’s interesting but totally sidetracks your search.
IMO it’s like Google 15 years ago where it was causing more panic and disinformation but people knew it back then - we laughed at Google’s persistent and incorrect diagnoses.
Now though, with boomers and young Gen Y and older Gen Z - maybe the dumbest yet most confident people alive it’s causing a lot of harm. Everyone thinks they know everything and because they inherit from their parents and are basically rich society rewards them for being retarded
I have started filtering for pre-2020 items when I am searching for articles.
Obviously not a silver bullet - for example, not much help if you're looking for modern tooling-related stuff - but I've found it really useful as a way of reading up on fundamentals.
I can't be the only one that scrolls past quickly the whatever Google-AI spurts out. If I wanted to ask an AI - I would. But I've learned that they hallucinate stuff with numbers or anything that's time sensitive/recent (even if that might not be the truth now). So I won't spend my time on verifying their output. I'm after a quick search, not a likely guess.
What's worse, when reading the Google outputs, I have learned to ignore almost any website besides a select few, one of which is Reddit. Which is very weird thinking about it, but it worked in the past.
The last thing I searched Google for was "india gdp growth" and Google gave me an estimate for India's 2026 GDP growth. It took like 2 seconds. It's almost certainly basically correct.
If I had tried to find an authoritative source I might have spent a minute on it. Would it have killed me? No. Would I do that for something more life-threatening? Of course. But flash results are... fine, for 90% of what I search for.
A fun essay and tickled my brain neurons.
----
Fundamental learning path does not change. Whatever searche result force feeds you it's ok. So long as you have a critical thinking mindset.
People simply read titles, and oh that's what happened.
If you apply the first order thinking, you can spot most of mistakes in docs and results. But then you gotta move onto the second order thinking to see if the conclusion you came up with will be good for the future.
Say, you searched for how to build a house yourself.
You see an AI result that recommends contractors.
But if you break it down why AI recommended the contractor, you can see harness (assumptions) such ass security (the builder/searcher should not be harmed), convinience (how fast do you need) without regards to your personal needs. - that's why comapnies want so much of your info, so they can try to guess better output based on your needs.
Now you breakdown using the 1st principle, you find out that you don't want those assumed AI answer, and you ask more direct question and guide it.
But then after applying the 2nd order principle, you might buy into the AI's recommendation that you don't want to spend too much time building one yourself for the safety (you are a pro athelet, who should not do heavy labor for the sake of the next game match).
Anyways, consider multiple thinking models and views.
Has anyone tried making a p2p search engine? Naively it seems like one of the p2p networks from the late 90s or early 2000s should be able to handle it.
Searches propagate between nodes, nodes specialize on specific sections of the index, which can be random + user preference (maximizing redundance and availability for the things people actually care about).
Soulseek already gets like halfway there and it's not even trying to be. How hard could it be?
So my MS research focused on P2P networks, and with the caveat that the area has been out of favor (and hence under-researched) for about 2 decades, I would say it would be extremely hard to build even decent quality search in a purely decentralized manner.
There were essentially two main ways of doing P2P search: flooding and Distributed Hash Tables (DHT). Flooding is exactly how it sounds: you flood a query across the whole network. The most popular applications like Limewire, Kazaa etc were flooding based, and as you can imagine, didn’t scale well. I recall some stories of havoc on networks due to the amount of traffic they generated. However they were simple and allowed somewhat complex queries, because each node basically ran a brute force search on its local index.
On the other hand, DHTs organize nodes in a distributed index structure such that queries can be routed in a deterministic manner, typically in O(logN) hops. DHTs were basically the granddaddies of distributed NoSQL DBs like DynamoDB and Cassandra.
So you can already see the limitations of DHTs: they need keys to be indexed, typically keywords. You had to pick a small handful of keywords for it to be manageable. This inherently limits the complexity and scale of the queries.
Both had an existential problem with spam. There were essentially no defenses against bad actors poisoning the network with spam and malware. It was a huge unsolved problem until Bitcoin popularized proof of work (which had been suggested in various forms before, but Bitcoin was the first practical, robust application.)
And neither could approach the sophistication of Google, which could extract very high quality signals from their centralized troves of data to refine and rank results.
Personally though I think the time is ripe for a renaissance. Spam and bad actors are still existential problems, but with vectors / embeddings the discovery aspect could be addressed. It occurs to me that the most popular large scale vector search mechanism is HNSW networks, which seem pretty amenable to P2P networks…
There's yacy. I didn't get useful results wheb I tested it though.
https://yacy.net/
Back around 2008 it didn't give great results either. It's amazing how long the project manages to stay alive though
YaCy[0] is a P2P search engine. There's public instances but you can self-host it as well. It includes a spider to do your own crawling. It's challenging now between CloudFlare and other anti-bot CAPCHAS crawling effectively is really challenging. Bots (AI and otherwise) have ruined the web.
[0] https://yacy.net/index.html
This might be one of the most ai written blog posts I’ve ever read
This is a very well trodden argument the poster is largely right about. But like others who have written about it like Nicholas Carr and Anil Dash, I agree with the concern, but it places a lot of agency on the technology and not the users decisions. On one hand, you can argue for a long time Google was the default page for the internet so you had no choice but to use it to search, on the other hand, people need to use their brains more when thinking critically. They need to learn how to use it to assist in thought. Yes, Google search results are ads, but scroll past the first few pages and dig. Also, refine searches and look at other sources by people you trust. Ultimately, this is a real concern, and I'm glad the author is updating thoughts on it.
> This isn't really anyone's fault in a simple sense
Or, in a more simple sense it’s entirely Google’s fault
Google's fault that SEO exists? The incentive isn't even there, they want people to pay for ads rather than gaming their results. But SEO finds a way.
AI;DR
It is a naive question but I do wonder how hard it is to scrape the web and build web indexes that are shared via bit torrent and self host a tfidf traditional engine using a distributed effort. How big the web really is?
Is there a lot of additional secret sauce that made google work well in its prime?
I don't think it hurts thinking (at least I haven't noticed anything on myself yet), but it hurts the internet a lot, and may even destroy it in parallel to LLM. Online search became completely unusable, and it started well before chatgpt arrived.
Now when I am looking for information on this or that I'm obliged to append reddit, Wikipedia or YouTube in my search to hope to have something with real content (and I hate watching videos to be kept informed and I feel there is more and more LLM generated content there also).
I even have a subscription to chatgpt now that I use more and more as a replacement for a Google search, or at least as a supplement.
I don't like it very much because in the best case I have only a handful of sources that it selected instead of being able to look by myself at all the possible sources, but it is still better than spending hours finding something that is not SEO-slop.
I'm convinced it will never improve because it helps google and the others push their AI stuff, and they desperately need it to justify the billions poured in it, but also because it gave them a much deeper control on what is provided to the users.
Yeah, I agree with what you are saying, people seem to take the AI response at face value. I answer lies in the users recognizing that HAL and others can, and right now often do, make mistakes.
I‘m outsourcing all tedious searches to ChatGPT now. Anytime I try searching anything non-obvious, it seems to be buried in the Nth page. Both Google and DDG have this problem. Chat spares me the brain power of sifting through the endless slop and grift, and I can easily click through and read the actual content.
Yeah they're not great at searching, but they make up for it by hammering 2,000 websites in parallel for ten minutes straight. Gets the job done!