Comment by pclowes

For me, Google search results have gotten so poor (and other engines aren’t any better), that I’d rather just ask an online LLM for what I’m looking for.

I was once very good at advanced Google search queries but they seem to no longer respect such queries - either showing irrelevant results or none at all (that should exist).

I don’t love LLMs, but they seem to not make up stuff very often these days and usually cite links to what they summarized. Sometimes the tone of the summary is slightly wrong “algorithm X was designed for Y” (when I know it wasn’t) but it’s otherwise very close to the mark.

What does amaze me, is the LLM seems to “understand” my question with very little context — I would have to give a human many more details about goals/intent.

I know damn LLMs are not capable of thought and are just a glorified search engine, but they do it well. Perhaps all my education made me little more…

I used to mock Sci-Fi movies where characters lazily dictated questions to the computer and it gave high quality answers.

We’re living in that world now.

LLMs-as-search-proxies have some pretty nice capabilities. For instance you can say "limit your search to scientific papers" and they'll do a much nicer job. I've also had some success recently prompting them with "I'm looking for reputable sources", e.g. recently I was looking for ways to repel deer from my apple tres. A naive internet search had vendors of shady crap jumping me. The LLMs pulled up relevant papers and university extension programs from my area.

Though I will say I get much better results from the LLMs I pay for than the free ones with Google or DuckDuckGo, which seem to be way way way more prone to just make crap up based on your search and cite web pages that, when followed, don't have the claim being made in the AI search results at all. By contrast every "source" link I've followed in the for-money AIs has 100% backed what the AI said it backed. Don't judge by the free AIs the search engines put out, those things are probably starved of resources and are nearly useless.

(Which I did not intend as a commentary on Google's plans here, but it is a data point of interest... that pressure to cut costs on the "free" services is quite directly at odds with providing quality AI services for the forseeable future.)

As someone who was getting information this way most of last year, I'm pretty sure I'll never want to again.

An increasing number of studies are indicating a reliance on "AI" leads to deleterious cognitive effects. I felt this acutely myself.

I've noticed a significant boost to my recall since shunning "AI" as much as possible.

The other day I found a comment here on HN and I wanted to know if it's true. I asked Gemini and here is the conversation https://gemini.google.com/share/2c1089ac6fd6

You can't do something like this with search.

I think LLMs are better at finding the most helpful sources now, but that's more a testament to how much the front page of web search has lost to low value LLM content.

Google has optimized their hardcoded search engine so much for the natural language searches people actually use, that they made it useless as an actual tool for someone who wants to find information. AI jumped over all of that and is BETTER at natural language searches, leaving the google search engine largely useless for anyone.

LLMs are so frequently inaccurate its crazy to think of it fully replacing search.

I've been trying to use LLMs for things and it makes mistakes all the time. Just this week i had multiple instances of various LLMs basically saying "just run the software with --flag-that-fixes-your-problem" or "edit the config and add solve-your-issue=true" hallucinating non-existant options. Even if i manually link the relevant documentation pages it will still just make basic mistakes. and if im having to read the documentation myself anyway to fix the AI's mistakes, why is the AI even in the loop.

its infecting search too, because blogspam/slop articles are managing to make their way into search results by just making up untrue information, claiming software can do things it cant, or has options that don't exist.

I'm not jumping in with both feet, either, but "never" is a very big word.

Search results are 80% SEO low-quality garbage nowadays. Very often, the sites even generate their content with AI. So for many use cases I stopped bothering with search and directly ask LLMs instead.

As a concrete example, some advertising supported topics place search as an unwanted middleman, may as well ask a LLM directly. Consider "chocolate chip cookie recipe".

Using google search, will return roughly infinite recipe sites. The sites were generated to spam AI generated recipes surrounded by advertisements. None of them are really any good because they were generated by a script and not looked at by a human until I come along and click. The standard is for all recipes to have at least 10-15 screenfulls of vertical spam wrapped by ads for recipe pages. The internet, at least using Search, is now useless for food recipes. I would have better, faster luck driving to the public library and looking in a physical cookbook; at least those recipes were probably tested at least once by humans unlike the advertising spam sites. Nobody has 45 minutes to watch 44 minutes of filler material surrounded by ads on Youtube either. If you want to cook food, the internet is near dead at this time, unfortunately.

AI search will plagiarize the "Original Nestle Toll House" recipe from the back of every bag of chocolate chips ever made. Its a good recipe and I've baked them many times over the decades.

I wish the internet were more useful, but the people in charge of it don't want it to be useful; here have some ragebait and doomscroll while watching the ads.

My google use is down because it turned to garbage. They're likely doing this because they poisoned their own well besides the advent of LLMs.

LLM hallucinations are better than Google results these days and I'm not even trying to tell a silly joke. It's more useful for an LLM to lie to my face about 10% of my query, be suspicious and dig out that useful information than to try to parse the absolute slop returned by a normal, non-AI Google query.

I don't comprehend how the average person gets any useful information out of Google.

That hasn't been my experience, I use LLMs in place of search. Two examples from tonight that I asked Grok for:

> Can you find the girl who did a bunch of posts critisizing David Graeber's Debt? I thought it was really well done

> I saw a comment on hackernews a while ago about the optimal amount of credit card fraud being higher than zero because of game theory dynamics, can you find it.

In both cases it turned up the exact posts I was looking for in like 30 seconds which would have taken me much longer using traditional search. I've had similar success looking for technical documentation. It's downright magical how they're able to turn my vague idea of what I'm looking for into a pointer to the exact thing.

Other way around. If I'm looking for the answer to a problem, I don't want the hallucination engine's half-remembered ramblings, I want the primary source that it's poorly attempting to reconstruct. But finding those primary sources has the potential to be easier, because LLMs effectively have built-in fuzzy search better than any classic search engine ever implemented.

In other words, I have no use for an LLM summarizer; I want an LLM librarian, working with me to say "beep-boop, here are some resources that seem relevant to your query, feel free to resume this session later if you'd like to further refine your search".

> On the other hand, search engines should be more like "<thing>s released 1908" or "<topic> and give you results talking about what you searched.

Is that useful enough to build a billion dollar advertising business around? My feeling now is not really.

Even for straight up searches, I find using an LLM to do a search and comb through the results is a better experience than Google is now for searching. If I'm specifically looking for esoteric web sites from 27 years ago on vintage computer hardware and software (thank god for Archive.org), Google is just ok for that.

> LLMs are 100% better because they'll give me an answer without trying to sell me their thing or pump me with ads.

Yet.

> they'll give me an answer without trying to sell me their thing or pump me with ads

Surely we all understand that any commercial model is going to inevitably metastasize into this.

Or subtly misrepresent politically inconvenient facts, or gently steer you into opinions based on a synthesis of broker data and demographic info, or quietly flag you in some database column due to exhibiting dissident-adjacent ideas or behaviors, or...

Yeah, they probably aren't doing (most of) these now, but it doesn't take much mental energy to extrapolate once you factor nearly every other tech company's ethical trajectory and the current geopolitical environment. Substituting classic search entirely with LLMs is not a savvy move.

My thought here is that there are many. They have proven to be commodities in most use cases.

As soon as one gets annoying, expensive, advertiser heavy etc. you just rip it out and replace it with the other one. AFAICT there is zero lock-in or moat. I often am able to switch models in one click or command. This is why all the LLM providers are desperate for a product layer/comprehensive tool set.

Sure maybe they all end up that way, but there’s plenty of reasons corporate customers will want private LLM usage that is not skewed towards advertising. I am happy to pay for that.

Also, open source models are a bulwark against another search style ad Monopoly.

Yes, but with an LLM can also say drop those links into a markdown file then use research agents to identify and stack rank the ones that are most relevant to this criteria. Add summaries and rationale for the ranking to the markdown etc.

This is a common flow for me and works with other skills such “as find recent PR’s in our code base that are related to this research topic”

Also yes, I don't care about the LLM and I am just using it to get what I want because that is what LLMs are for.

That is incredibly weird, yes.

I find it extremely useful. So much so that I am annoyed when I have to use it in a handicapped fashion such as on the phone.

It makes me wonder why like 90% of the apps on my phone exist. I just want everything to be markdown files, skills, MCPs/API and then a nice TUI or voice to text.

Because they will still try to pump ads via AdSense. That's why I've created a platform called Zero Ad Network, where developers can monetize their sites by NOT showing ads and get paid by the platform subscribers.

Wait, you guys don’t think Google AI responses will be geared for advertising? It already is FFS, massively.

In my experience, LLM usage follows an exponential distribution i.e. most people using LLMs are not using many resources, but a very small few are using a massive amount of resources. Most LLM usage is trivial and tiny, most of what I use LLMs for could be done by a local model with a decent web retrieval tool. What some people I know use LLMs for requires massive volumes of tokens. Its that small power user base which I would bet they are targetting.