Hacker News Reader

Given how Search-Engine-Optimisation (SEO) has been gamed, what will make you think that somehow this NEW system, that's really prone to prompt hacking & already promotes sponsors' products over alternatives, won't be?

For me it doesn't need to be a perfect, bias-free information source (no such thing exists). It doesn't need to solve all my problems. It just has to be useful in certain contexts, and I will use it while also trying to be aware of its limitations and conducting my own "sanity checks" to make sure the information can be trusted.

Of course it will be gamed. Everything can be gamed. Like you said, search engines are also gamed; that doesn't mean I refuse to use search engines.

I guess the pragmatic answer is that you don’t need AI for that. You need good filters. I don’t like Zalando one bit, but I’ll grant them that it’s easy to find the right clothes on their website because they have very good filters.

LLMs don’t ‘know’ if a pair of jeans is a tapered slim fit with a gusseted crotch, at least not by default. But if the brand uploaded them as such, the filters will find them.

That’s just a quick take. I’ve tried to shop with LLMs and the results are mediocre at best. Of course search, filtering, and content tagging could always be improved, instead of “just slap AI on it”.

It's worth pointing out that in the Googlebook video she's not asking Gemini to shop for her. She starts with a picture of herself, and asks Gemini to combine it with vintage clothing photos from a website she was browsing, to help visualize what she would look like in those clothes.

This is more than just search + filtering. I've done similar things when trying to visualize home improvements, and find that it really is a useful way to help validate my ideas.

So far a lot of the negative responses I've gotten have been along the lines of "only a fool would let AI do your thinking for you". But I find that it's a useful tool sharpening my thinking. Brainstorming, overcoming my own personal biases and gaps in my knowledge, idea validation, etc. Like "rubber ducking" [1], but the duck actually responds with some pretty insightful advice with surprising frequency.

Do I "need" AI for shopping? No, of course not. Can it reduce friction and lead to more informed buying decisions in certain cases? In my experience, yes.

Of course I've seen plenty of useless "just slap AI on it" jobs, too. Netflix put out an AI chatbot that I found particularly egregious, for example (I think maybe they've taken it down since). I didn't find Amazon's "Rufus" to be very trustworthy, either. And I know I'm coming across as pro-AI here, but in other matters I have plenty of serious concerns about AI. I'm just hoping to have a more nuanced conversation than "shopping with AI? Gross!". Or "only a fool would use a product built by greedy corporations!"

[1] https://theconversation.com/stuck-on-a-problem-talking-to-a-...

I don’t have a side on this particular fight. I’m using LLM-powered coding assistants to help me with a side project, and I enjoy DuckDuckGo’s AI assist thingie. All good. But I would prefer better filtering than “slap AI on it”. On the whole, I very much agree with you.

Nobody is picking their laptop for the best AI integration. You can do those things just as well on every other platform. In fact, additional AI integration is universally a turnoff to most normal people.

Uhm, the only reason I bought a refurbished laptop last year (an M3 Max with more RAM than I've ever had before) was to run models locally.

If Google was launching a new laptop that was meant to run models locally I would be really excited.

The key phrase from GP is "most normal people." You are, with love, a particular kind of freak. As are we all.

“Nobody” is pretty universal, yet there are plenty of people now buying new computers specifically for their local AI capabilities. It’s not all Mac minis, some of us prefer laptops.

By "AI integration" we are talking about UI, not running local models. See the link this thread is about.

My terse answer to, 'Why not use AI to shop for [X]' is that if you are letting AI do the shopping for you at any level, you aren't actually distinguishing products by features or quality or it's ability to solve a problem. You are being fed junk that is likely paid to be moved to the top of the list.

It's probably a nice feeling when you can put in a list of soft requirements to ChatGPT et al and get a list of things it recommends, but I would suggest you are a fool if you think those listings aren't bought and paid for.

In an era where the gap between a 'good product' and a 'bad product' is growing ever larger and the price is not an indicator of anything, the onus to actually become knowledgeable re: "How to identify products worth buying" is becoming greater and greater. If you are using AI to do the shopping for you, not only are you not building that muscle, you are actively weakening it as a chatbot convincingly recommends something to you based on unverifiable platitudes about 'quality' and 'value' - a recommendation that was, again, bought and paid for.

So yeah, that's gross and I would argue pretty strongly that it's just as brain rot adjacent as something like Tiktok. Like Tiktok though, I expect it will see at least some level of popular use, and also like Tiktok, I think it'll end up making the population dumber on average.

> I would suggest you are a fool if you think those listings aren't bought and paid for.

At no point in the process did ChatGPT direct me to any listings. I fed it my criteria, and it gave back a text response listing car models that met my criteria. The only links it included were links to reddit posts and other car reviews. And the results were useful to me because they pointed out where my own pre-existing biases had caused me to overlook one model that I probably should have paid more attention to.

What you are suggesting feels more like a potential future threat than my actual experience thus far.

I found the listings by conducting a separate search on a used car listing website - and the number of matches that met my criteria were small enough that I was basically able to look exhaustively through all the matches. But shopping for used cars can be a little confusing at times because there are a lot of different configurations that change every year. Sometimes the listing might just say something like "2022 Touring, Safety Package" and include a bunch of photos - and identifying whether a given listing has a particular feature you are looking for requires some investigation (ideally they would include a full list of options, but often times they don't). Or often times the listing itself might contain incorrect information. And I found ChatGPT to be a useful tool for quickly making sense of the various configurations, and of course conduct my own sanity checks to be sure the information is not hallucinated.

I'm not suggesting you should solely rely on AI for shopping (although in some cases for low-risk purchases it may be fine) - but rather as an additional tool to aid in research and decision making.

> What you are suggesting feels more like a potential future threat than my actual experience thus far.

Do we really have to litigate this? Have you been on the Internet at all in the last 2 decades? Do you seriously think that even if that kind of advertising vector isn't being paid for today, it won't be tomorrow?

It is almost childishly naive to assume that these companies that are bleeding billions will have the ethical fortitude to say 'no' to Chevy / Ford / Jeep / Whoever when they offer them a check to make sure Toyota and Honda are unceremoniously just de-prioritized as recommendations.

---

Beyond that, the issue is still that you are not going to get complete market coverage. It's feasible that you might on certain smaller market segments (Cars, for example), but something with much more producers and products in the segment has no chance. You would be better off spending the time to understand the market, what differentiates the products in it, and how to think about the parameters involved - all things that are being just abstracted away by asking a Chatbot for a list of requirements.

> Have you been on the Internet at all in the last 2 decades? Do you seriously think that even if that kind of advertising vector isn't being paid for today, it won't be tomorrow?

Conflicts of interest are nothing new - dating back to newspapers, radio, television, and search engines. And yet in all of these mediums companies have figured out how to display sponsored content while still maintaining the trust of their users. AI companies have a similar vested interest in maintaining their users’ trust (not to mention adherence to current and future advertising regulations).

> Do we really have to litigate this?

Yes, if you are going to assume the worst possible outcome, then you must also explain why other outcomes - such as clearly distinguishing sponsored content from “informational” content - are not possible.

> Beyond that, the issue is still that you are not going to get complete market coverage.

Which non-AI information sources promise complete market coverage?

> You would be better off spending the time to understand the market, what differentiates the products in it, and how to think about the parameters involved

I agree this should be the end goal in decision making. And in my experience AI can be a useful tool to get there.

> all things that are being just abstracted away by asking a Chatbot for a list of requirements.

ChatGPT doesn’t just give a list of results without context. It’s also quite good about justifying why it gives the results that it does. And you are free to ask follow-up questions, and fact check the responses against other sources.

I treat it the same as basically any other information source that I come across. I fully understand that it is not perfect. But that doesn’t mean that it’s not useful.

I can very clearly imagine it always going for branded products where brand is not required, unless specifically prompted not to, which the average person won't do.

> I need dishwasher tablets

Could mean buy a 30 pack for £25 which have all the marketing buzz surrounding them, or buy the own brand 45 pack for £5 which does the job just as well.

> being fed junk

As if the products you find in mass market brick'n'mortar stores are any different.

Or the information I would be fed if I walked into a car dealership and asked a dealer. Unbiased information has never been a thing, and while AI introduces a set of tools along with a new set of risks, it doesn't really change the fundamental problem of needing to vet your information against trusted sources.

They on average, are. That's kind of my point.

Yes, if you engage with the 'designed marketing channels' for products, you will end up with junk. If you want to have stuff that isn't junk, you need to do some leg work. A chatbot will not do that for you.