Hacker News Reader

I now only google when I need to sometimes verify that what chatgpt is suggesting isn't BS and I'm immediately reminded that google is barely better between SEO content farms, SEO SO scrappers, legacy information from 10 years ago. And that's with ads blocked.

Between chatgpt+ for general guidelines and copilot for specific implementation details, programming feels very fun and alive. And I'm very skeptical to subscribing, but chatgpt provides so much mental relief getting some answers immediately that I'm ecstatic being able to use/pay for it.

I still cant believe google thought they could ruin the internet like that and continue to dominate search

To be fair, it has worked well for them for at least half a decade. It’s only now they are paying the price.

Greed is a hell of a drug.

The response speed of GPT isn't fast enough for me to want to use it like Google. Even with plus.

The response speed of Bard is much better in my experimentation. The creativity of the output on Bard is lacking, though.

That's good to hear. I was concerned it might just be a hard limitation in the early days, tech wise.

Can Bard output quality code like GPT? I'm in Canada so couldn't use Bard sadly.

If you ask it to "write code to do X", then it will say that it can't write code. I think this was hardcoded.

But I asked Bard: "Can you implement fibonacci in Go", and it outputted valid code. And then I asked it "what if I wanted to avoid recursion", to which it replied with valid Go code that used a for loop. But it also suggested me another very bogus way of doing it, by using a "function pointer", which was very bad. F(x) would output x.

So, don't expect ChatGPT level of quality just yet, but I think it will get there pretty fast.

I've found Bard to be very responsive too. Though you could argue it's not getting the same amount of traffic ChatGPT is.

That said, I don't expect Google to rest on its laurels. Sure, OpenAI is executing swiftly these days, but I think they've been prepared far too long for this. I expect Bard to become much better very soon... then will the AI wars truly begin.

Even though Google returns results instantly, for most things I still have to evaluate and click links and skim them for the information I want. Sometimes I have to do multiple searches (like one for retinol and another for beta-carotene in the following example).

Yesterday I heard about retinol (vitamin A) mentioned in a nutrition podcast. I know carrots are high in vitamin A, but I didn't think they had retinol, so I wanted to learn more about that.

I whipped out my phone and asked GPT-3 "retinol vs. the vitamin A in carrots" (something I know you usually can't ask Google).

A few seconds later, I learned that retinol is vitamin A's final form in the body, thus you get it directly from animal products, and beta-carotene—found in plants—is a precursor to retinol in the body.

I do these kinds of searches all day. One thing faster about GPT as well is that I don't have to consider the "query engineering" to make Google return what I want, I just ask GPT a question streamed from my consciousness.

Yeah for questions with absolute answers that can be summarized like that it's perfect. Basically what Wolfram Alpha offered with quant data.

Although my partner is a lawyer and she sometimes asks it to summarize cases (without providing the full source material manually) and it sometimes invents entire details in the cases in a very persuasive way. So you always have to be careful and double check if it's important.

Code and case summaries couldn't be much different. With code you get compile time validation, run time validation, (hopefully) test validation, and you can generally look at a block of code and say "seems like this should work" or "this makes no sense". You get none of that with facts of a case.

SEO garbage is such a problem these days that, if I were Google, more than using AI as a new frontend for search, I'd be trying to find a way to use it to defeat SEO.

OpenAI should sell a feed of AI generated responses for search engines to consume and flag AI generated content.

SEO for AI will be worse. Their output is a probabilistic distribution conditioned on the text is has read, so you can place text that heavily biases the output. Usually figuring out what this is can be kind of hard, but since OpenAI lets you query GPT-4 for credits, you can just mine what these special phrases are.

just wait till chatGPT starts including ads in results.

not just 'including', but 'integrating'. A chill ran down me when I read this and realized the awful inevitable truth of it.

My personal opinion is that this has nothing to do with SEO. All those developer-focused blogs serve Google ads, while the official Go documentation does not. Google is trying to boost revenue by promoting results with ads.

This is exactly it. Google is trying to get people to stay on the pages they click, and they optimize for this behavior, not for utility or for quickly accessing information.

If you go to a page and immediately find what you are looking for, Google interprets that as failure.

This is why Bing and AI based search can completely beat Google web search, because they are trying to serve the searcher rather than serve ads.

The problem AI search has is monetization, openAI can't run gpu server farms to serve 100M customers for free forever. This was a problem that plagued early google as well, sergey and larry were against using ads for monetization, but in the end that i what drove the business to profitability

I wonder if AI search could be sth consumers would happily pay a tiny bit for

They wouldn't pay for AI search, but might for AI that also searches.

I don't get where this "Everything must be free+ads' mentality comes from. Netflix thrived for a decade using subscriptions, Microsoft and Apple's businesses are primarily subscription based.

AIs are so extremely useful to some users, that they'll gladly pay $$ per month to use it. And the size of 'some' increases radically with each leap in AI's capabilities.

GPTs don't have to be gimped to just search. Even GPT-4 today with wolfram alpha can obliterate most low-level tutors. It will be a personal tutor, therapist, coding assistant, secretary, storyteller, entertainer all in one very soon. People will pay for it.

sth?

"Something", perhaps.

Yes :-)

…for now.

AI search seems to have similar dynamics to traditional search. The people writing the stuff you’re searching for need to make money. (For example, AI doesn’t magically know about real-world events. It needs a human being to write about it.) Then the company offering the search product needs to make money too.

If you compensate sites for using their data in your AI product, then there will be the same incentives as now to game the system (AISEO?)

If you don’t compensate sites, then they’re not incentivized to produce the content that makes your search product useful.

I still find it hard to believe. I've been at Google until 2015, which was a long time ago, but at that time the separation between Search and Ads was considered near sacred. Ads representatives couldn't even ask "Hey, my customer asked why their site is below this other irrelevant site for this query, this seems like a bug in search ranking..."

Pretty much everybody understood that we will lose in the long term, if we let ad revenue even be a consideration in search ranking. (...which makes today's bad search experiences even more puzzling.)

I'm in neither Search nor Ads, but have been at Google long enough to have seen the massive change in internal culture between 2015 and the present day. I don't want to speculate on any specifics, but it's pretty clear to me that we shouldn't be applying 2015-era logic or perceptions to today's Google and expecting them to hold.

I joined in 2014 and left in 2019, and it was much different company by the end. My wife still works there, and she tells me it is different still from since when I left.

Developer focused blogs rarely show google ads. Upsells to saas like fly.io are more common

I just don't think this is true. I just think Google has become actually bad at search, and there is no malice.

The problem with SEO spam was always there (remember expert sex change?) but they beat it before... but now it became bad again.

But they have monopoly anyway (bing is even worse for actually searching, ddg is just bing, brave search sucks) so they don't really care. What are you gonna do, use Ask.com?

pay for kagi.com, although that's not an option for >99% of internet users.

I guess the problem is that Google doesn't have a metric for "clicked, wasted ten minutes, and was annoyed."

I think you could actually get them to accept a new strategy with open arms, where everybody needs to have some Google watcher process installed that checks and reports back what you do with the search results (I don't think unless the target website imports Google JS scripts too it can't track any more?) and more specifically, your body reaction to see how you feel. "It's necessary for better search results!"

Right, they just care about the 10 minute captive audience bit and how many ads they displayed in that time.

They could have let Chrome users rank any page they're browsing to get quality feedback. The installed base is massive.

That would just be gamed like everything else. Downvote the competition, pay SEO folks to rank your stuff high, etc

It would be quite costly to pay millions of people to do your bidding when you're competing against millions of real users.

This has been my thing too. I work in Networking and sometimes I see interesting stuff, but between cisco's beefy docs and obscure blogs I have no idea what the actual use case for it, or how the SD-WAN magic wand does the thing. ChatGPT will get me to verifiable information so fast, and give great in-context examples. It's renewed my passion for the field just when it was getting dull.

> I get all these developer-focused, SEO-optimized blogs that are kind of helpful but often not, and I have to scroll all the way down to get the official docs. I really wonder if that has to do with AI-generated content becoming more ubiquitous

No. There's a cottage industry of content writers that just churn out this stuff by the bucket.

I have tried to hire writers in the past that should've been niche experts but just regurgitated content like this, and had SEO experts trying to convince me that we should actually aim for low quality because higher-quality articles will be too dense for people. (both contractors got the boot and the "content" went into the bin)

I know what you mean, and it's kinda annoying. I often click the blog links knowing I'll probably be disappointed when I see they're certain sites. Personally hosted blogs tend to be a bit better.

I wish they added some kind of categorised search. So you can add "+documentation" or "+blog" to at least narrow it down if you know what you're looking for but can't remember the URL

They used to have blog search and it was crushing when they dropped it. I'm sure usage was low, but the few visits blogs got from that mode presumably dropped off the earth after that. Surely identifying personal blogs and the UX cost (tucked away in the Tools menu) couldn't have been too costly.

I'm confused.

Surely ChatGPT, etc have been trained on content from these very same SEO optimisation hellholes you're talking about if it was trained by scraping the web?

Is there some trick that allows it to determine whether the content it's being trained on is SEO garbage?

Maybe I'm missing something here, but it seems like all you're getting is an amalgam of SEO content feed to you by a bot...

Not an expert but one of the datasets used to train OpenAI GPT is Common Crawl and they use Harmonic Centrality not PageRank.

https://www.searchenginejournal.com/harmonic-centrality-page...

Did ChatGPT really write Go code with generics for you? Last time I fed it some code that implements generics it said, "this is invalid syntax, Go does not support generics" which would make sense since the knowledge cutoff date is 2021