Hacker News Reader

LLMs have helped us keep HN up and still (relatively) fast amid a huge onslaught of new traffic over the last year or so. We've been able to solve problems and find opportunities for optimization that we'd never have gotten to without the help of Claude or ChatGPT.

Although this is one of those threads where people love to report how much they dislike HN, how it's a poor shadow of its former self, and how they barely look at it anymore—a popular local sport for 15 years or more—but to the extent anyone still reads the site, there must be some positive impact in having it running and rendering pages, even if you don't like what's on them!

Hopefully HN is just getting more popular, but isn't a large part of the "huge onslaught of new traffic" driven by AI companies scraping the site for content or writing comments for various reasons?

Fixing the problems created by LLMs using LLMs seems like a bit like hiring a bully for security to prevent that same bully from beating you up.

But yes, there's plenty of positive things that are now possible. See as a combined net-positive vs net-negative, I'm still leaning towards the net-negative.

I fear this is not the only reason even if it is true. The more negative reason is a loss of skill/talent. We used to be able to know what to do in these situations, and how to fix things when they went wrong. Companies like Y Combinator used to be able to hire talented systems engineers who knew how to minimize the risk of things going wrong, and fix them when they did.

Now with LLMs it is just vibe-coded until it is fixed. Find the right prompt and hope the statistics is in your favor. If this is the case, there is nothing positive about it.

I take your point! But the positives are still positive. I've been amazed by how much LLMs have helped us do.

None of it was undoable otherwise, but we (where by "we" I mean "I") would never have gotten around to it otherwise. Lowering the bar from months to days, or weeks to hours, is significant.

that new traffic likely was from AI and their scrapers. So no, not much of a net positive.

I mean, scraper traffic should be pretty easy to tell.

Now, AI posting bots are much more difficult.

I think the scrapers have gotten pretty good and can imitate real traffic. And it isn’t just scraping for model training. A large amount would be LLM tool calls searching the web (and HN) for info about some prompt.

I don't think that AI has to influence every single person's daily life to make a huge impact. I think for probably 90% of people it's pretty useless out side of being a better search engine. But that doesn't mean it won't have a huge impact on science and technology at large. There are a lot of snake oil salesmen trying to stick it into everyone's face, but I don't think that should be conflated with the real progress that's being made.

I would say 90% of people do not yet realize how useful it is. It is not useless to them, they just don’t understand yet what it can do.

And there were a few years there where a lot of BS AI hype turned them off to it. So a lot of people rightly dismissed it and don’t realize the inflection point it hit in the last few years.

ChatGPT work is basically a personal assistant for $20/month

> ChatGPT work is basically a personal assistant for $20/month

A personal assistant that will happily tell you that you’re the world’s largest supergenius who invented perpetual motion. I wouldn’t want that if they paid me $20/month to use it.

Current Opus and Fable simply don't do this. You post is stupid.

There’s more than 2 LLMs in the world. Now whose post is stupid?

It’s never told me that, but it does help me fix problems. For instance the other day I noticed that my state unemployment insurance rate had increased from 0.75% to 12.75%. I absolutely could have spent hours of my time scouring the website, trying to call and waiting on hold, talking around the issues with a poorly trained state agent.

Instead, I asked ChatGPT what happened. It did a dive and figured out that I was being assessed a penalty rate because I had switched to a new payroll provider that does not file wage reports in my one quarter (seasonal business) where I have no wages. Without a wage report, you are considered delinquent.

You have to love the government, it is the only place where you can be delinquent on a payment you do not owe. ChatGPT filed the correct wage report, navigated to the Support site requested a retroactive rewrite, which was granted and saved me many thousands of dollars.

Would I have done this myself eventually? Probably. But I work a lot in season. Something more important would have been neglected. It would have taken me an amount of hours that I would happily buy back for an entire year of ChatGPT business. It didn’t even hit my 5 hour usage limit to do it.

So maybe instead of asking it how it feels about you, ask it to do useful things. It’s my employee, not my girlfriend.

>because the technology needs to find it's way into applications and appliances.

I hear what you're saying, but speaking for myself: can we *not*? I really have absolutely zero interest in having ChatGPT on my microwave or refrigerator. In fact, I'd rather have my appliances *not* be connected to the Internet, all things being equal.

>I'd rather have my appliances not be connected to the Internet,

It's getting harder and harder to find things that don't connect to the internet as time goes on. The moment a manufacture figures out how to sell your appliance data for more than it costs to add it, seemingly every model they produce gets it added with the only choice of going to some other manufacture if you don't want it... and even then finding one not internet connected gets nearly impossible.

In general most people don't care enough for people like us that don't want connected devices to move the dial.

What are you doing in tech that has not been impacted by AI?

i just gotta say that i find it hard to believe that this comment is true for you but you hang out on hackernews.

almost everyone here is a software engineer or works in tech, they all use ai all day to do their entire jobs.

> even thought the benefits aren't all that clear yet

it isnt 2023 anymore. wtf are you talking about.

>he hype will die down, especially once we start to see the actual costs, e.g. ChatGPT doesn't actually provide enough value for the average user to warrant a subscription,

again, it isnt 2023 anymore. none of this is true.

> they all use ai all day to do their entire jobs

I assure you they do not, many truly do not care and don't want to. That's just what the hype and the article and the stories tell you. The HN bubble tells you that's true, but 90% of my co-workers don't even want to touch Claude, despite being given a license.

And again, if e.g. ChatGPT is providing so much value, why aren't people buying?

> why aren't people buying?

They have "more than 50 million consumer subscribers" and "More than 9 million paying business users" [0]. That's for example more subscribers than for Adobe Creative Cloud.

[0] https://openai.com/index/scaling-ai-for-everyone/

> more than 50 million consumer subscribers

So, about 0.6% of the world’s population is willing to pay for something that’s supposed to be hugely useful for literally everyone (as opposed to Adobe products, which are only for digital artists)?

Not everyone in the world has $20/month of disposable income, nor does everyone have the disposable time to learn how to make it hugely useful to them.

Also, these are just the numbers for OpenAI. A lot of others are ok with e.g. "paying" for it by going to Google's AI Mode, and getting some ads.

> nor does everyone have the disposable time to learn how to make it hugely useful to them.

True, I’d rather spend my time to get useful skills whose usefulness cannot be destroyed by a single corporation deciding to pull the rug.

> A lot of others are ok with e.g. "paying" for it by going to Google's AI Mode, and getting some ads.

Since everyone who uses Google the search engine is forced to use Google AI, we can’t get meaningful statistics for those.

So ChatGPT cannot deliver you $20/month worth of productivity to offset the cost? That's at least my point, for developers, businesses and a ton of use cases, access to a good AI is worth every penny, because it easily pays for itself. For the average consumer, that's just not true, apparently.

HN is in an extreme hype bubble around LLMs, because of AI coding. Normal people outside of the tech bubble range from 1. not caring at all about AI to 2. disliking or opposing it.

Meanwhile, here on HN, you have people who earnestly say things like "AI is the subtrate the future runs on"[1] (actual quote).

1: https://news.ycombinator.com/item?id=49417949

Just saying "it isn't 2023 anymore" isn't an argument. Yeah it's not 2023. And LLMs still can't actually do anything useful outside of sift logs and write small scripts. They still hallucinate things regularly so that you have to check everything they do, which negates the time savings they supposedly provide. And that's not even getting into the externalities! LLMs are a very crappy technology.

I guess Hex-Rays and Vector35 must be hallucinating their effects for, say, reverse engineering. OFC rev isn't be all and end all of computing but the effect LLMs had on security research and adjacent spaces (say, CTF) is interesting

https://github.com/HexRaysSA/ida-nexus

https://sidekick.binary.ninja/

> they all use ai all day to do their entire jobs

I am so fucking sick of this. No, you dumbshits -- ALL of us DO NOT use LLM's to write code. Just fucking STOP assuming every developer needs the same crutch that the rest of you can't seem to live without.

I use it a couple times a week, not even daily. It's useful about half the time, and a complete waste of time the other half. It's a useful tool to reach for sometimes and I'll keep using it for the things it's good at, but it hasn't really changed how I work. Shrug.