Hacker News Reader

> I definitely need LLMs for one critical thing: search that works.

Hmm, that seems weird.

> Google has become clogged with outright spam and endless layers of indirection

So.. the problem isn't search, it's that Googles ad business has destroyed googles search and created spam to try to get ad clicks.

I switched to Kagi a few months ago and I'm much happier. Most of the blog spam is just gone, and the things that slip through I can nuke.

yeah I should try it, I'm just a laggard.

Still though, the way that LLMs can give customized answers to complex specific questions, questions that probably don't have any specific page on the web, feels like a leveling up beyond traditional search engines.

here's a recent example: I asked how to call a particular function in some library from a little-used language, and it not only told me exactly how to do it but wrote the FFI wrapper. I could definitely have dug that information up but it would have taken a lot longer and then I'd still have had to pore over documentation to write that tedious code.

"Can give" seems not super relevant honestly.

If you ask an AI for historical pictures of Hitler shaking hands with Zulu fighters, you might get them, but you'd be a fool to think that's facts. Text is exactly the same. The tech is exactly the same. The result is exactly the same. People are just so used to text being about facts via school that they are fooled, but they know about paintings from before kindergarten.

Your example isn't about facts or search though. In the case of code we can easily check the result. It's not a case of facts really.

I wholeheartedly agree (that first ask from ChatGPT, then double check with whatever, and that Google results have become junk - mostly). Alas, I can totally see how AI companies will monetize this by allowing advertisers to influence the answers. And not in a add-in-a-sidebar fashion, but baking the advertisers product etc. into the answers. Personally, I find this somewhat disturbing.

Indeed, I think this is a key area where regulation is needed. If answer from AI is influenced by 'something unexpected' - what ever that might be, then it has to be clearly highlighted in the answer.

I think you missed the point of the article.

It's not meant as a criticism of ChatGPT and some LLMs, more like a criticism of corporate drones and "thought leaders" that latch on latest hype instead of fixing their core problems.

Oh I probably did because none of that is interesting to me. Every 5 years another crop of omg corporate drones are so awful consciousness appears, but once you've been through 3 or 4 cycles, everything new is old again. The rants aren't more interesting than the suits are. (well, - the rants at least have their fun side)

What interests me is that there is new technology here that does something relatively magic - that just doesn't happen often. Corporate crap and MBAs and suits and founder hucksters are forever with us. I'm sure it was equally true decades before me and you showed up.

I'm rather annoyed that this is so, because now I have to slightly get off my ass and learn something, so as to reach a better state of lethargy later.

I'm a ranter by character, but you know who I'd pick between the ranters and the drones? The drones. The drones are the 100 million sperm racing to fertilize the egg. They're all losers (save one infinitesimal winner) and it's easy to point that out. The ranters stand on the side saying "those fucking idiots. those morons. what losers they are. as opposed to me, who sees this." But the ranters are the dead end, smart as they may be. The sperm are at least in the race.

This it's incredible if you keep in mind it's just gonna find the most "common" words that come alongside your query in some jobbled together answer

Machine learning isn't just a word frequency machine. If that's all it was, we'd have this technology decades ago.

I'm really wondering about this. Is it just "recycling things people have said before", or is it genuinely generative?

"Recycling things people have said before" is basically search, and that is hugely valuable and quite enough for me. If it's genuinely generative, that's a cosmic leap beyond search.

My guess is that it's not genuinely generative, but rather that the long tail of "everything that everyone has said before" is so vast as that it feels like magic when it's retrieved.

But LLMs have shocked me enough on the search front that I'm no longer smugly confident about this.

Have you seen the Othello paper? [1] To me it really puts paid to the idea that LLMs could just be stochastic parrots, in the sense of just rearranging things they've seen before. They at least can apply world models they devised during training (though whether or not one does for some given prompt can be a different question).

[1] https://thegradient.pub/othello/

I mean, if the question is if they act as just pure, 100% search and nothing else, the answer is pretty self-evident.

I'm not much of an LLM user, but the few times that I did turn to it for programming advice was in rare and obscure situations that weren't really discussed anywhere on the internet (usually because they contained multiple issues in one). The LLM tended to produce something I'd call a reasonable answer, especially on topics that weren't completely obscure.

But we don't even need to go that deep to answer the question. For example, if an LLM was pure search, you couldn't make one generate text in some specific style or with specific constraints, unless that exact answer already existed somewhere on the internet. They can mash up ideas or topics, and still output good or reasonable data.

The billion dollar question isn't whether it's generative - it's whether the generative capabilities are "enough". Machine learning is about finding patterns, and a complex enough pattern finder will be very good at approximating answers accurately. LLMs don't actually have an accurate "model of the world" learned - but they have something that's just close enough on certain topics to make people use it as if it does.

The issue with google is true. That being said ai isn’t the answer in my experience. Wikipedia is a better source of casual information I find. And if I want to dig further I look for the real sources if I can, not the slop written by journalists but the actual material people in the field are engaging with. The primary sources. The documentation. If I can’t find that or understand that material, then thats that and I move on, considering anything ‘accessible’ is going to be an example of gell mann amnesia and not worth my time.