I Quit Google Search for AI–and I'm Not Going Back | Hacker News Reader

I Quit Google Search for AI–and I'm Not Going Back

wsj.com

41 pointsBostonian77 comments

I'm wondering why, in this day and age, it is still difficult to search for a specific type of jeans, with a specific size, style and color, and availability. And filter it by shipping method/cost.

Same for other products. E.g. resistor 0805, 1/8W or better, etc. etc.

Why does every online shop have their own (often broken) filtering system? Why can't the information be gathered by an AI, and turned into a super filter?

Sounds like you are talking about Amazon? Which is a front end for online shops.

Amazon's search is somehow even worse even though they already have a lot of the information in a structured format.

I suspect that's by design.

For the same reason some department stores have the up and down stairs/escalators on opposite sides.

https://ux.stackexchange.com/questions/49307/why-do-departme...

I'm talking about the web, not just Amazon. Amazon's filtering is limited too. I want a universal filter with many options. And not just shops that are captured in Amazon's sphere. And my search engine should have no incentives other than providing me with good search results.

Because most products are in their own ecosystem with no overarching standard way to share information. You have some standards but usually for specific niches, no product FHIR.

Just making sure every product has a unique ID would be a revolution. https://en.wikipedia.org/wiki/Global_Trade_Item_Number

Google’s Froogle was a bit like this. Obviously they had to cancel it.

People talk about the web being corrupted by ads, but it's important to recognize that it completely fails at informing people what products are available to solve the problems they have, at what prices, from which vendors, and how to transport them most efficiently. Instead it's like walking through an unregulated flea market full of pickpockets and everybody's asking you to sign documents.

Yeah, everyone is talking about free markets. But no one likes the idea of a transparent market.

I guess they talk about free-to-scam-you markets, not about free-to-move-and-see markets.

It's not the web that fails. It is the vendors who fail, which is their own individual fault. Any vendor who takes online sales seriously can have tremendous benefit.

Amazon has terrible filters for that type of thing.

Probably for the same reason you can't ask in a Target whether the Wal-mart up the street has it in your size - the individual vendors are incentivized to avoid you going to another source, and hope you'll decide to go with something "good enough" rather than paying the cost of going to another place and comparing things.

That cost is much lower outside of brick and mortar travel times, but not 0, especially for people on crap internet connections or slow computers, and people tend to just want to get it done rather than shopping through N sites for the best option, so they pick "good enough".

That, plus the vendors don't want you to think they're just an Amazon frontend so they have their own UI, plus if they allow suppliers to populate the item fields, good luck getting the suppliers to provide consistent results...

> the individual vendors are incentivized to avoid you going to another source

Yes, but a search provider would not have that incentive. It can simply scrape the shop websites and parse it with AI, then turn it into filters.

Only if that AI does not hallucinate, otherwise it would be useless.

If you immediately find what you want you won't browse around and likely buy something with bigger profit margin or just more stuff.

It's not good for business to provide a good user experience, correct information and so on.

Google had "Froogle" specifically for shopping. Not sure what happened there, as it looks like they renamed it to "Google shopping" (https://www.google.com/shopping) but it just redirects to broken google search page.

https://en.wikipedia.org/wiki/Google_Shopping

I feel that

I was looking for braided BSP flexible stainless hoses in 3/8th of an inch with female to female ends that were longer than 1m

Boy I've had to scroll through pages of 1/2 inch hoses and female/male ends because fuzzy word search is completely unfit for that stuff

You can't even get Amazon to give you the right results based on filtering items by title. I was searching the other day for lithium rechargable AAA batteries and it gives you a mess of mostly AA batteries. Including ni-cad.

Interestingly eBay does a better job now of guiding sellers in filtering items into the proper categories, at least for the products I shop for.

Amazon is a dumpster fire.

For some products, this exists. I use Geizhals/Skinflint, which is technically a price comparison site, for pretty much every purchasing decision. If they support that product category (they're mostly focused on tech).

But if they do support it, you get all the filters you could ever ask for. This is the online searching experience the 90s promised. e.g., https://skinflint.co.uk/?cat=hdssd or

Yes! But the point is that nowadays an AI should be able to collect all this information for all product categories, and provide a uniform UI. But indeed, this is the kind of filtering I'm thinking of. It would also be nice, if, on the product page you can easily search for alternatives. I.e., all the filtering options should be presented for that product, and you should be able to make modifications and then hit "find alternatives".

AI is a heuristic like any other. You don't want heuristics for this, you want actual data.

Manufacturers should just be required to provide a standardized data sheet for their product (as they already are for e.g. energy consumption data in EU), which could then be automatically mapped. Then you just need stores to list the barcode for each product, and you can map everything automatically.

Still not complete. Try searching for Elitegroup in the category Mainboards there, and see what you get.

It seems like they closed shop, or stopped distribution.

Which isn't the case at all.

That applies to several other brands in that category.

Extrapolating from that to other categories I know less of, I assume similar omissions in coverage.

Why that is escapes me.

Maybe there is a fee to be paid, and they couldn't be bothered?

You are incredibely right, I recently buy 2 jeans 514 33/34 ... and they are not the same size as thus bought one year before ... Seriously Levi's are you drunk?

Drunk on cost savings from moving production from factory to factory around the globe.

I can think of a couple of reasons: manufacturing tolerance, differently sourced / dyed fabrics (make sure to wash your clothes before first use).

There is also something called vanity sizing (https://en.wikipedia.org/wiki/Vanity_sizing) where sizes go up without the numbers going up, so that people still think they're the same size.

If you're buying clothes a year apart, they're likely using a different supplier using different grades of fabric, differently calibrated machines, so the clothes won't feel or fit the same.

The most annoying thing, and reason why I don't "search" for clothes online, is that all the (web) store results are US based. What's the point of getting results in the US when I don't live there?

1) AI isn't that smart and doesn't search the Internet in real time, nor can it distinguish between real and counterfeit products. There's a significant liability issue if BigCorp AI sends visitors to IP-infringing shop sites.

2) Search engines would rather show you irrelevant results than an empty page. This is true of Google, Amazon, Netflix, you name it.

3) The more custom a product is, the less likely it is to actually be in stock. I've seen plenty of cool products on Chinese electronics stores, many of which have quietly refunded my order after determining it wasn't in stock and wasn't coming back.

That's a pessimistic view. I think with even small effort a lot of improvement can be made here.

There will no doubt be some techbro startup that promises to solve this: after all, who wouldn't want to build a platform that guides you to these products and take a cut of every sale?

Whether or not such a platform actually meets your desired specifications is another matter. The money in this industry is in affiliate marketing and ads, not in helping people find stuff.

I can see where this is coming from but if your view was correct, then companies like Kagi would not exist.

Because nobody benefits from providing that information. Having choice isn't good for sellers. They want to sell you their product, not allow you to check whether their product+shipping is the cheapest.

When I wrote my first webstore I thought the product data would be reasonably static but I've gradually learned it is more like a first person shooter.

AI could still gather it but one would have to accept that all parts of a product may change which may create an undesirable game.

I think we’re close to the solution. The problem for the 2000s was getting everything online, 2010s was getting APIs setup and I hope the 2020s will be when we build truly useful filter engines.

Google Shopping probably comes closest with its Vision Match feature: https://www.google.com/shopping/page/visionmatch

I don't think I am there yet. I got too confident about AI and was lazy to do some search instead or think through the problem solution. I wasted my 3 hours by not doing so. LLMs are getting better but search and stackoverflow still helps.

Doesn't sound like the LLM was the problem here. By your own definition, you are lazy.

People quit Google Search to avoid "search-optimized junk" just to use AIs that answer using "Search" lol. Look closer at the sources your AI answers give you. They're not always pretty.

Every time I try an AI search engine, I always end up going back to Google. Their AI summaries at the top of search results are, honestly, "good enough."

This is, in part, because most of my Google searches are 1-4 words; as opposed to the full prose that I need to type into an LLM.

What am I missing?

I find LLMs are useful if I don't know enough to know what to search for; if I need to craft a "long-form" search query; or otherwise need to learn something in a Q-and-A format.

Edit: One of the things that ChatGPT has replaced Google, for me, is when I need a quick joke image. Instead of searching through Google Images, I just think of a quick prompt of something silly. I'm using some silly ChatGPT images in some test scripts; and now that the US courts declared they aren't protected by copyright, I know there is 0% chance I will get in trouble for them.

> What am I missing?

You are missing the conversation. In ChatGPT, Perplexity or Claude you get to have a conversation on the topic, instead of a one-shot answer.

But the conversation is time consuming.

IE, it's like having a meeting when a simple email or IM will suffice.

Google is deciding for the user that a IM will suffice. In reality a meeting is needed very often, and Google search doesn't have the option.

No, I decide when I need a conversation or a simple search. Google doesn't do that for me.

No you don't. Google Search only offers search, and one-shot AI. You have to go somewhere else if you want a conversation.

I know in advance if I want a conversation

I think code monkeys or journalists etc. have GREAT experience with LLMs.

In case of niche and more technical knowledge demanding task, LLM just fails. Eg. in my case, reverse engineering. It just spits BS all day long.

Probably business types too. If you ask "what's the best email sending company" it will give you the most popular ones and all the right talking points to convince your underlings to switch. Now the salesfolk don't have to bother with golf games anymore.

A few months ago our product manager copypasted into a feature squad chat the results of asking an LLM to give some options for encoding route-navigation data for transmission from a server to a client device (both running our software). It gave about ten options ranging from half-decent but over-engineered (e.g. geoJSON) to bizarre (e.g. OSM, rinex, nmea). He didn't let on that it was an LLM response - the implication was that we should work through the list and anlayse the best one. But as both sides were using our own software so we could just use our own JSON encoding. Fortunately the head of engineering was also in the chat, immediately realised what was going on and dismissed the demand. Of course the real issue is that the PM is overreaching, but such behaviour is enabled and/or catalysed by LLMs. I don't doubt that similar things are happening in other companies that don't have the privilege of assertive leaders pushing back. The wasted effort could be significant.

Honestly if it’s not web (front and backend), it can’t figure it out. It makes sense though. Web has millions of examples. RE and other niche fields have far fewer examples, and the examples that do exist are extremely specific. How to pop a shell on device X: one example, extremely specific. How to write user auth for a web app: thousands of examples and generic enough it’ll work anywhere.

Its early days but for our B2B SAAS, we are seeing an uptick in traffic from Chatgpt and other similar AI tools. I even have prospects sharimg their prompts as i always ask them. Google search is going to have serious issues in next few years if they continue on their current path of literring page 1 with garbage "sponsored posts".

As opposed to 'ai' littering all of the response with more subtle ads that aren't even marked?

Yes

https://archive.ph/3EqIe

I find this a little worrying from an ethical point of view. The suggestive power of web-based LLMs could be far greater than that of traditional search, and this could well be commercially exploited if we're not careful. With traditional search, providers like Google throw in ads that look like results and bias the genuine results based on funding. This is dark, but can be circumvented with some basic guidance on how to interpret the UI, and the web pages themselves are not affected. LLMs provide a wall of text which all appears reasonable, but could very well be biased or manipulative in highly subtle ways based on whichever political parties or corporations have paid the most for influence. Whilst in principle you can tell users to externally verify the claims, people will bother less and those external sources will become less accessible/useable as LLMs become more widely used. I hold out hope that they can be mitigated, but I don't yet see how.

This article uses an LLM to summarize itself into three key points. Those three points are milquetoast, making me much less interested in getting behind the paywall and reading the actual article.

Maybe this is a value add. Maybe the article is milquetoast and should be skipped. But either way it’s an example of how LLMs are not just competing with search they’re degrading the reading behaviors which make search more valuable (details, specifics, exact match).

It is an underrated point though that LLMs are competitive with Google search because they remove anti-user ads and SEO spam. In the long run LLMs will have that same problem though, because it’s a business dynamic not a technological dynamic.

I've been kind of expecting to see the LLM impact on Google's 10-Q for a while now.

Even if LLM queries are just a small fraction of overall internet queries, I suppose they're stealing eyeballs from higher-value searchers.

High value searchers for Google are people looking for car insurance, which is not something anyone is doing with LLMs.

For "search engine" tasks I have switched to Kagi.com and find it more useful than Google's results. I am not using it for shopping but to find original content that offers multiple perspectives on topics I am researching. I find Perplexity helpful when I am trying to get a basic overview because it provides links to substantiate the content. Your mileage may vary.

Microsoft is barely even mentioned - a major defeat for Satya. During its entire existence Bing has been waiting for an inflection point. The inflection point came and Bing failed to take advantage of it. Perplexity is what the new Bing should have been. At this point it is fair to say Microsoft has squandered their AI advantage they got through early investment in OpenAI.

It’s shame Google search is so terrible because everyone using AI versus visiting websites is just going to exacerbate a dead Internet. There’s not a lot of incentive to generate new content that nobody will see. And with AI relying on this content, it seems like a pretty shortsighted feature.

Last night I asked a question in Google and also in DuckDuckGo, using their AI assistants... One assistant said "The answer to your question is yes" and the other said "The answer to your question is no".... So yeah, not a thrilling result.

Mind sharing the question/prompt? It'll be fun to try it in other LLMs.

As long as ChatGPT uses Bing for searching the web, Google Search is not going anywhere.

Critical thinkers avoid LLMs. They make too much work for us. So much fact checking!

Citations and reasoning models eliminate this to a fair degree. Deep research cites most of its conclusions. We may not be fully there yet, but we're getting there.

I'm fine with LLM as long as it sources its responses. It is true however, that LLM interpretation of the source links can be wrong.

The problem with search engines like Google now is that they stopped giving people the results they want, but it often shows things Google thinks people want, or whatever SEO trickery Google stopped trying to fight (especially when the links are full of Google ads...).

With Google’s recent constant popup about switching to the app in iOS Safari, I’ve pretty much 100% dumped Google for search. I use either DDG or ChatGPT search.

The popup to me screams coming desperation.

100%. The app begging is terrible, Reddit-level. Thank you for the reminder about duckduckgo.

I left big tech on the other hand. I use RSS for most 'news'.

I have bookmarks for most common Internet places: https://rumca-js.github.io/search

I have my music bookmarked also https://rumca-js.github.io/music

I have some parts of the Internet crawled. It takes a long time to boot, but once project is running I can search all domains: https://rumca-js.github.io/internet

Some people complain about AI slop. I do not encounter it.

That does not mean I do not use google, or chatgpt, but only for things I cannot find at my bookmarks.

I use Google's AI often by mistake. Even when I know it is wrong I will quote it here on HN and give credit to Google AI for the misinformation for posterity sake. Historians will want to see what output it was giving people especially if there should be any cases of public manipulation through these bots. I do not know where else to archive it. AI is still lacking fact-checkers. Google AI still can not provide any references to where it learned the misinformation often referred to as hallucinations and as far as I know there still isn't a team that prunes the misinformation and I have no idea how to report it or downvote it.

I use open ai and it does, in fact, provide references. (I turn on web search)

I was referring to Google's AI since that was the topic at hand. Nice to hear that OpenAI has references.

It's extremely sad we have to resort to this, but I found that it completely solves the problem (where applicable, _obviously_) to limit the search results to before 2022. Especially for images, with the en masse pollution of the stock websites it's impossible to sort through the slop anymore.

I use AI for search as well, specifically the ones like Phind that perform several their queries for you and parse the sources. You still have to double-check the contents of the sources because the AI will hallucinate. But on top of that, as others have said, sometimes the only sources it can find are the same SEO-optimized AI slop that Google serves you as the 1st results

Unless I need a map or want to see a picture of a celebrity, I use ChatGPT for pretty much everything.

I try to use different things depending on what I'm searching for. For techy stuff relevant Reddit group or stack is often better. For things that you end up with Wiki, their home page is there anyway to search. For news related search it's a problem with any of them as they usually show paywalled results.

Reply on news.ycombinator.com