Hacker News Reader

Comment by disambiguation | Hacker News Reader

Everyone: telling me how great AI is.

No one: making anything great with AI.

Sourcegraph: an AI company, routinely promoting their LLM-optimism blogposts to HN, perpetuating the hype cycle their business model depends on.

View on HN

This is the true litmus test IMO. If LLMs are so great and make everyone so productive, then where are the results? Where are all of the amazing products being released that otherwise would have required 10x the investment? Shouldn't there be _anything_ we can point to that shows that the "productivity needle" is being moved?

I don't think it makes everyone so productive, tbh! If you really know what you are doing and are willing to burn tokens, it will really take your work to the next level—provided you don’t work with a niche product or language that the models weren’t trained on.

If I may use an analogy, it's like what sampling is for music producers. The sample is out there—it’s already a beautiful sample, full of strings and keys, a glorious 8-bar loop. Anyone can grab it, but not every producer can sell it or turn it into a hit.

In the end, every hype train has some truth to it. I highly suggest you give it a shot, if you haven't already. I’m glad I did—it really helped me a lot, and I am (unfortunately, financially) hooked on it.

then find a "great" artist willing to cosign that they made something using that AI, vs their own talents.

The demo is the work. Social media views is where the productivity is being moved. And VCs are paying for this ball to keep rolling hoping to eventually cash out big some day.

AI is an irrational market at the moment and this is not going to change anytime soon.

What are you on about? It's happening everywhere, in all companies. It's literally impossible to miss and how to handle the massive increase in code output is becoming a well-known problem. I think it's interesting how the naysayers claim that it's not happening, yet simultaneously worry about the large amount of low-quality code being produced.

lines of code is not equal to features released. especially when you have to diagnose faulty code in production later. So... low quality code is worse than no code.

This sapling is twice as large as it was a week ago, which was twice again as large as it was the week before. Why, at this rate, it'll be bigger than the whole world in but a month.

I remembered the dialogue from the movie Snatch:

    T: What's happening with this sausages, Charlie?
    C: 2 minutes, Turkish.
    -- 5 Minutes later ---
    T: How long for the sausages?
    C: 5 minutes, Turkish.
    T: It was 2 minutes, 5 minutes ago.

I don't know why I remembered it. Is it AI, or self driving cars, or both. Huh.

self driving cars are here, just unevenly distributed. Waymo operates in several cities already, providing millions of rides to people. Just because you haven't seen them doesn't mean they don't exist.

Waymo operates in tiny parts of several cities when you look at the maps. They didn't finish the first 90% when you look from that perspective.

Waymo provides a tech demonstrator in a closed circuit, and you can pay to ride their demonstrators in select areas of select cities. It's not half bad, let's be honest, but having these limits puts them to a point very far from "self driving cars are here". Even Tesla isn't there with their camera only subsystems and "if you crash you keep both pieces" policy. It's another tech demonstrator you can buy.

If we accept your point that "self driving cars are here", we should accept that Fusion energy is a solved problem because WEST tokamak held plasma for 22 minutes, electric propulsion is a solved problem since we have electric trains for a long time, and Hyperloop is fully operational because I have scale model in my backyard which carries thousands of ants every day.

People open their phone, order a car, take a ride from an arbitrary point within the geofenced area, to another arbitrary point inside that area. Within that area. they're handling real life conditions. dogs and children running into the street, bicyclists, moving trucks blocking streets, construction. Other cars. Maybe that's not enough for you, but that's not a carefully staged tech demo with a car driving a loop around a fixed track at Disneyland that's gonna break the moment it has to deviate beyond what you're allowed to touch. Unless you think Waymo is paying all the drivers in LA drivers to drive different around their cars, I guess.

If the tokomak was energy positive for those 22 minutes, there was more than one of them, and they were running it multiple times a day then I would say it is. But it's not. I'm not sure why you have this need to lump them together. Fusion is quite a different problem, with yet to be solved materials science problems and control systems issues. I'm not claiming that fusion's here, and also I don't think quantum computers are gonna be here anytime soon.

Don't forget the electric air taxis!

Drone-bot delivering my new dishwasher-top-rack-rails today; waiting 4 days for a human; SMH

Where's my TacoCopter??!

https://tacocopter.com/

It's the opposite. Most people are not are boasting about their productivity improvements but it's everywhere. Unless you work at a company where you're not allowed to use these tools it should be impossible to miss. Even the most hardcore naysayers I know are now using AI tools. The new discourse is whether the massive increase in code output leads to issues or not (I think it does), but claiming it doesn't happen is not a serious take anymore.

> We’re talking about each developer gradually boosting their productivity by a multiplier of ~5x by Q4 2025 (allowing for ramp-up time), for an additional amortized cost of only maybe $50k/year the first year. Who wouldn’t go for that deal?

OK, I'll take the other side of that bet. If in Q4 '25 devs using cursor or whatever are 5x as productive as me using emacs, I'll give this AI stuff another chance. But I'm pretty sure it won't happen.

You should probably compare the same dev using AI vs not using AI. Otherwise you can use that argument for anything.

Notepad is a better IDE than emacs if we compare a really good dev using notepad vs a shitty dev using emacs.

Ok, if I notice other devs suddenly getting 5x as productive I'll give it a try, but so far no such effect has been demonstrated. It seems like a pretty straightforward research question, and you'd think if there was any demonstrable effect the companies selling these things would use such research to market their products. So where is it?

There's a dead body lying there but you won't believe they're actually dead until someone creates a video reenactment of the killer stabbing them and shoves that video in your face?

Speaking of videos, https://youtu.be/opi1s_5Dm-c is probably below your level and it's kinda long.

I don't get it. What is the dead body thing about?

There are people building things and making money using these new techniques, but because they haven't stopped to produce research studies and white papers about their work, you want to presume it doesn't exist.

Yes, without evidence there's no effect. It's telling that nowhere on cursor.ai's website do they appear to claim using their tool will actually make you, or crucially your team, more efficient.

evidencthe absence of evidence doesn't mean there's no effect

There is no trustworthy evidence that I'm aware of to support the claim that "using an LLM IDE makes developers 5x more effective". Do you disagree? Testimonials don't cut it, because individuals are prone to hysteria, delusions, and just making stuff up.

All I'm asking for is a few trustworthy case studies showing unambiguously that it's worth my time and money to pay attention to this genai stuff. That shouldn't be hard for a trillion dollar industry, right? Like show me this isn't just hysterical hype. Don't tell me, show me.

What does it even mean to be 5x more effective? Because that's the thing. Testimonials is where we're at because how do you measure "more effective"? I'm not making any claim on a number, 5x 10x? .2x because you have to go back and fix the shitty code it wrote? We don't have a good way to measure programmer productivity. We all know lines of code is ridiculous. And how do you account for how much brain space it takes up. If the LLM lets someone bang out useful code in between meetings all day, vs needing some truly quiet time for some deep deep work to properly get into the right zone... how do you metricize that and show, what, lines of code and interruptions per hour on a graph to have a line that goes up and to the right?

It turns out that it's actually a really hard problem to measure, so, even for a trillion dollar industry, it is hard.

What did you think of that video?

Oh good. When I saw the graph I started wondering what I was being sold.

Reply on news.ycombinator.com