Hacker News Reader

Comment by noah_buddy | Hacker News Reader

You have touched on something I am incredibly excited about. ChatGPT is going to do away with a lot of the intro level questions you ask researching certain sorts of technical problems, especially as it becomes better able to understand images. ChatGPT removes the need for you to ever open a manual (unless you want a deep understanding).

How do I turn off this feature in my car?

How do I replace the fan in my fridge’s compressor?

Why is windows repair not working in this scenario with this PC?

This tool is going to be incredibly deflationary in many services geared around repair when now it can tell you almost exactly what to do and soon it’ll be able to produce videos or images of each step on demand.

View on HN

> and soon it’ll be able to produce videos or images of each step on demand

As long as they're not hallucinated media...

Most of the Googles SERP results already are delusional or hallucinating spam, so the bar is low and for me at least ChatGTP is less bullshit.

Yeah, when the “site:” search hack doesn’t get me an answer and I fall back on the normal SERPs, most of the time the top several results are all:

“Authoritative title that references my exact search” -> page body restates my question, gives an oddly reworded summary of Thing in Question #1 -> “some people online have said” followed by snippets lifted from the ether -> repeat for Things 2,3,etc -> “our suggestion is to look at the features and decide based on what other customers/users are saying”

It’s perfected vapidity wearing the skin of an editorial review site and takes a LOT more cognitive load to suss out the phoneyness compared to the last time the SERPs were packed with spam (~10 years back).

hallucinations don't invalidate the product itself. I really hate to see this constant lazy refrain on here.

Its just like anything else. Trust but verify. Some will just trust, others never will... People probably over-trust the tech now, but just like wikipedia it will be approached a little more carefully as people learn its shortcomings.

Unless you're trying to perform open heart surgery or doing something intrinsically dangerous, its probably going to be additive. I wouldn't trust it to help me assemble a warp engine, but it may be able to give me some decent pointers for guitar technique or how to change my oil.

It’s not a lazy refrain, it’s a serious downside of the technology. It’s tiring and stressful to supervise the work product of an assistant that is extremely capable sometimes, but a compulsive bullshitter at other times. Just as it’s tiring to supervise an “autonomous” car that often deftly navigates the road, but sometimes wants to plow into a school bus.

I think that the concern about LLMs making up information are absolutely valid, and a serious concern. At the same time, I think most people focusing on those concerns aren't thinking much about the alternatives, namely Google.

For many people, Google results have gone from being an absolutely amazing demonstration of what could be done if one had the resources to crawl and process the entire internet to a complete waste of time at best, clogged with spam and nonsense and misinformation. I've been using Duck Duck Go for years now, and recently my middle-schooler switched to Duck Duck Go because I got answer to a question she had already asked Google before checking with me. Some people even pay for Kagi search, because Google is just that bad.

So LLMs don't have to be perfect to be better than the alternatives, not even close.

> Some people even pay for Kagi search, because Google is just that bad.

Small correction - from what I have heard in our community, most of our users pay for Kagi search, not becuase Google is just that bad, but because Kagi is just so good. In other words we are making the value proposition be "best search in the world", one worth paying for.

(Kagi founder here)

Reply on news.ycombinator.com