Hacker News Reader

Comment by college_physics | Hacker News Reader

"In order to change an existing paradigm you do not struggle to try and change the problematic model. You create a new model and make the old one obsolete."

Transcribing this profound insight of Buckminster Fuller to the current debate: You cannot out-do google by inventing "smarter" search. You need to create a new business model that makes the old one obsolete.

Is there any evidence that all this algorithmic magic is enabling new business models that are not based on adtech more likely? It would be a blessing (presumably, because things can always drift further into evil) but, so far at least, there is little to point to such an imminent disruption

View on HN

Google beat their original competitors — Altavista, Excite, Yahoo!, and so on — with an incremental improvement, not a new model. At the time (2000-2003), their search was better and faster, certainly, but not several orders of magnitude better.

In fact, Google was something of a reactionary model at the time, by rejecting the push for bloated, captive "web portals" and going back to the simpler, stripped-down user experience that Altavista had succeeded with originally.

Given how bad Google's results have become, I think you can certainly outdo it. Anecdotally, it's amazing how often I find myself reaching for ChapGPT these days to find basic facts that Google can't. I think Google is right to be afraid.

Pagerank (TM) was a Google innovation and the reason Google had much better results than the others and it took many years for the SEO spammers to catch up. That revolution, combined with the philosophy of the speedy, focused homepage and the elimination of irrelevant results by always using the ‘AND’ operator was why their popularity exploded.

At Voila, a subsidiary of France Telecom, a study around 2005 found that their search engine offered nearly the same answers as Google, while not in exactly the same order.

Despite this objective comparison, in users' view Google's results were much much better.

I remember a German CS professor telling me eyes in eyes with obviously a lot of schadenfreud "Google will kill you".

And I answered "No they will become as us, big, fat and ugly".

> At Voila, a subsidiary of France Telecom, a study around 2005 found that their search engine offered nearly the same answers as Google, while not in exactly the same order.

Order is a pretty big part of tbe value prop of a search engine, so ”nearly the same” results but “not in exactly the same order” is not as much similarity as you seem to be implying.

Come on, there's no absolute correctness for a result in being in Google's first versus second answer. The actual order depends a lot on the user's context.

Even Google doesn't change the order of answers based on the user's context. Otherwise it wouldn't say in Google Web Console "your site appears in nth position"

> Come on, there’s no absolute correctness for a result in being in Google’s first versus second answer.

“Absolute correctness” is an irrelevant standard. Order can affect satisfaction without "absolute correctness" being a thing that can even be assessed meaningfully.

> Even Google doesn’t change the order of answers based on the user’s context.

Yes, it does. [0]

> Otherwise it wouldn’t say in Google Web Console “your site appears in nth position”

I think that reflects the experience of a no-context user (e.g., not logged in or otherwise tied to a history to provide context.)

[0] https://en.wikipedia.org/wiki/Contextual_searching#Automatic...

All a Google killer has to do is resurrect Altavista's near operator which is what ChatGPT is doing in an abstracted way.

Other people invented pagerank and used it for pageranking before Google. It's just a centrality metric.

Jeff Dean even posted on G+ about how he essentially independently invented a variant of pagerank at DEC, but his supposed collaborator couldn't figure out how to open a tar archive, and that was the proximate cause of his leaving DEC.

Actually Google got a patent for the concept of Pagerank, number US9165040B1.

You mean stanford. Getting a patent in no way means you were first.

Google is the initial assignee and a patent provides exclusionary rights to the inventor of a technology. Great if someone else thought of it earlier but no one published anything using it or thought to get a patent, because otherwise this very important patent would have been void and it wasn’t.

Robin Li's rankdex patent predates pagerank, see https://patents.google.com/patent/US5920859A/en?oq=US5920859... as well as the pagerank patent, which cites it.

Patent US9165040B1 does not cite US5920859A, I’m not sure what you mean.

It wasn't just incremental. It was a massive improvement! I still remember how floored I was with the quality of results compared to AltaVista which was my main search engine at the time

Depending on the day of the week, I might be doing a standard enterprisey dev project or I might be doing a “DevOps” project automating some things around AWS (where I work in ProServe) using the AWS SDK.

Before ChatGPT, I would spend a lot of time searching on Google for the right API call and the response structure:

Problem: I need a simple Python script that list roles in an AWS account that has at least one of a list of command line specified policies.

Before ChatGPT: I would look up the api call to list the roles (the API service area of AWS is huge), then look up the API call to list the policies for the role and then look up the API call for the “paginator” that pages through all of the results since the API call only returns 50 results at a time.

After ChatGPT:

“Write a Python script that allows me to specify a list of AWS IAM roles that contain at least one of a given list of policies”.

I get “working” code and then I start polishing it.

“Use argparse with required parameters and use a paginator to iterate through the roles”

Again expected results.

“I need the list of ARNs printed out as a comma separated list”.

I copy the code into VSCode, test it, and I’m done.

Later on for another project, the customer prefers JavaScript. I copy the code into ChatGPT, and tell it to convert it into JavaScript. It works perfectly and is idiomatic JS along with proper async/await handling

Lol dude I just googled “script list IAM roles that contain policy” and this was the top result. Probably less typing than the usual silly authoritative tone people use when composing an “AI” prompt , like they’re Tony Stark or Captain Picard .

And trust me, once u google stack overflow enough pretty soon you won’t need to ask anyone how to output a csv

https://stackoverflow.com/questions/66127551/list-of-all-rol...

that’s true. But that code snippet didn’t meet all the requirements. I laid out. And it definitely was less typing using ChatGPT once you add all of the requirements I laid out.

And then there is the context switch. This is just one little script as part of a larger system where I’m designing an entire system. It’s much easier if I’m already writing out a long set of documentation, diagrams, slides, etc just to open ChatGPT and keep writing English.

Not to mention when you’re juggling a couple of projects and one is Node and one is Python.

And just in case you are trying to gatekeep, I’ve been programming for a few years - let’s just say my first hobbyist code was in 65C02 assembly language in 1986

I agree completely. Hilarious example. But, hell, if a product's pitch is "you'll feel like Tony Stark or Captain Picard" it will probably be successful.

The entire thesis is that ChatGPT can give better results than Google.

ChatGPT gave me a specific answer with the command line arguments that I specified and allowed me to specify a list of policies and output the result in the format I needed

Google gave me a code snippet that I had to modify for my use case.

Exactly. Search ads are pretty much the perfect internet business model - if my biz is having a sale on lawnmowers, who better to target than people googling for "buy lawnmower"? If this is done right, everyone walks away happy and google makes a killing. ChatGPT is great, but it's pretty far from providing good answers to "where can I get a good deal on a lawnmower right now?" They might take over a decent share of esoteric, "school assignment" type queries, but they're not really a threat for the high-value searches.

If ChatGPT had up to the minute crawl results and my Geo location, that'd be enough.

"Help me buy whichever budget electric range Wirecutter recommends at the moment".

> Okay, here's the store near your home that is selling the Frigidaire electric range that Wirecutter is currently recommending, at the best price.

> Would you like to buy it online at roughly the same price or get the address of the store?

I don't understand how this is different from what Google currently provides, except that it skips one step (navigating to wire cutter, selecting some text and right clicking, then "search web for [product name]").

The only difference, which is not an improvement, is that you've suddenly given Wirecutter total power. At least if you make users visit the site and read a bit of the review, the reviewer has the hurdle of writing some text that can convince a human reader that the results are not totally bought and paid for.

Right! I think most people are uncomfortable with totally outsourcing their decisions, however smart the service is. I want to do my own research, and feel like I've gotten a good overview of the market before I commit.

You're saying if ChatGPT had a regularly updated search index, and was capable of effectively ranking webpages to duck around SEO (adversarial webpages looking to hack the rankings), it'd be competitive with Google?

Can you explain how this is any different than what Google does?

Not GP, but, and this is a hypothesis, it seems to me the realtime scrape might be enough. GPT seems incredibly intelligent in figuring out the ranking(!), importance(!) constraint-search(!) and personalization(!) by itself. Meaning, if OpenAI can do realtime fine-tuning of the web on a daily basis, they _might_ have sidestepped ~25 years of R&D from Google, and be an order-of-mag better.

When I google, I have an internal dialogue about whether the results are shit to begin with, if they aren't, maybe the thing I'm looking for is more niche, so it's not in the top ten results. ChatGPT could for example just maybe try 5 google searches in the background, just glean data from the description, title, question asked, date, etc, and maybe use some of it's own scraping on the top 10 results to reorder them and give better summaries.

E.g. imagine when you search via chatgpt. It googles : search phrase, search phrase + reddit, search phrase variation 1, variation 1 + reddit, var 2, etc... all in parallel almost then uses that to mini-tune an ai model to find the stuff that best matches what you're really wanting to find, and then scrapes top ten for results.

Ads work because they're embedded in interesting or useful things. No one would use a search engine composed of 100% ads.

Chat-GPT can be used for this: returing interesting results. The ads don't have to be served through chat-GPT, but could be given as a side result.

> No one would use a search engine composed of 100% ads.

Isn’t that pretty much what the Yellow Pages used to be? As long as you know what you’re getting…

Or virtually all trade magazines, Auto Trader, swingers' magazines, plenty of hobby magazines. There are tons of things you buy for the ads. MaximumRocknRoll in the 80s and 90s, the reason you bought it was to find out where to buy records, the articles were extras (and they told you what records might be good and what bands might be good to see.) Their coolest co-publishing thing was issuing guides to all of the clubs and other vendors in every city and state that touring bands needed to book.

These are all ads. The enemy isn't ads, it's an extremely intrusive and adversarial advertising industry posture.

Ads are useful, they tell me where and what things are available. The problem is that everyone uses the methods originated with patent medicine to sell everything.

The methods of selling patent medicine are simply the methods of confidence men; the "medicine" was probably just some unholy mix of alcohol and mercury. Patent medicine marketing emphasizes secrecy, because they're usually mostly frauds; and also because to any extent they are not fraudulent, they want to be exclusive and not copied. Patent medicine marketing also emphasizes leads, surveillance and secret dossiers. When you're selling dreams, you have to know what people are dreaming about; you also have to know how much they have to spend.

Our advertising problem is a legislative problem. We find it very difficult to police this kind of behavior because it is so widespread among the powerful (and is usually what got them powerful.) It is also completely rational behavior because the methods of con men will sell more regardless of the quality of the product i.e. a good product sells even better using underhanded methods. So there's simply no way to protect your world from being run by the best con men without creating legal obstructions for them.

Yellow pages were not 100% ads. They were where you got phone numbers and addresses for businesses, which was important pre internet. They did have tons of ads alongside the listings, but the unpaid listings were core to the offering. If you wanted to call for a reservation at a particular restaurant, check movie showtimes at a particular theater, you would go to the yellow pages, and to the regular listings specifically (not ads, although they’d certainly be there trying to entice you elsewhere). (You could also get the numbers via 411 but this cost money, 50 cents or a dollar IIRC).

> If you wanted to call for a reservation at a particular restaurant, check movie showtimes at a particular theater, you would go to the yellow pages

If you wanted to find the number of a particular restaurant, the white pages were far more efficient.

The yellow pages were primarily useful when you wanted a list of restaurants, not a specific one.

Something else that was nearly 100% ads: the old Computer Shopper magazine. As I recall, they printed just enough editorial material to qualify for the "literature" postal rate rather than the "advertising" rate.

My memory is that originally business listings were excluded from the white pages, added in later years. Been a while though!

> No one would use a search engine composed of 100% ads.

And why I'm using Google less and less. I've literally had the entire first page be all ads.

You are describing a comparison shopping search/aggregator.

If you are having a sale, you are going to be near the top of the results for any reasonable sorting scheme. (An actual sale, not here's 50% off of my bullshit price)

And searching Google for "buy lawnmower" is a good way to buy something you didn't want or need. Even if you actually needed a lawnmower. The only people it helps are those who don't care whether the item is good, and don't care about the price at which point they might as well just walk into the nearest store and buy the first one they see.

Search existed before Google. They also didn't invent adtech. Instead of a marginal improvement, they made it so much better that the previous option looked silly. If such a disruption happens again to the core technology (e.g. search through a ChatGPT-like entity), whether the financial backing is adtech will be irrelevant.

You know how better app for finding good restaurant can't find good restaurants of there are no good restaurants?

Search is dead, spammers killed it and Google was happy about it because it has become an answer machine where they can say the answer is whoever pays the most. Greatly deteriorated experience from what it used to be.

Now someone builds an answers machine with much much better user experience with all that data which Google can no longer surface.

I don't know if Google can quickly match that function but even if they do, are they going to be able to replicate the experience if the answers machine choose the answers by a bidding war?

> You cannot out-do google by inventing "smarter" search.

Pretty sure you can, because Google search sucks balls. (First and foremost because it ignores context. E.g., searching for programming API documentation is tuned entirely differently compared to searching for movie recommendations.)

That said, Google is not in the search business, because there is no such thing. Google is in the context advertising business.

Buckminster fuller was notoriously unsuccessful at changing anything, but was very successful at making it seems like he would have been successful if it wasn’t for all the conspiracies against him stopping his success.

Taken in that context puts it in a different light.

While I agree with your post, it is somewhat funny that not one of Fuller’s ‘disruptive’ products ever made a commercial dent.

Is there any evidence that all this algorithmic magic is enabling new business models that are not based on adtech more likely?

I think the revenue assumption here lies on the fact that GPT generates content when the web wait for the user to provide the content. Generates congent, for lawyers or hospitals for example - pardon my lack of imagination. So a basic business model would be to charge per request I suppose.

I remember that hope.

I'm using GPT3.x to make money at work

Reply on news.ycombinator.com