Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

GPT-4 is phenomenal at Code

github.com

Post reply on HN
Print view View original

211–220 of 221 posts

Re: GPT-4 is phenomenal at Code

#211

Post by hwayne »

ozten wrote: post #26

I don't know TLA+ and haven't had time to learn it. Could GPT-4 be a boon for esoteric PL that have had trouble gaining mass adoption? Clearly, one needs to learn and understand TLA+ before one can trust generated proofs, but maybe GPT-4 could be the mentor or gentle on-ramp I was missing.

I hope so! So far I've been unimpressed with ChatGPT and Copilot for generating TLA+. It mixes in idioms from other programming languages, like RETURN, which make no sense in TLA+.

What I'm hoping is that it'll be helpful for writing J. That'd be a feat.

Re: GPT-4 is phenomenal at Code

#212

Post by lisasays »

drusepth wrote: post #185
lisasays wrote: post #149

I'm the "really good", but definitely not "phenomenal" camp. That is, it works - but what is with the repeated class declaration for every h3 element? 1. Pruning The whole point of CSS is to have to declare properties like this once, and only once, for a group of functionally similar elements. Also note that switchTab has to be told the position of the tab it's switching to (it should be able to infer that the elemen…

They're using TailwindCSS for this, which is full of one-liner utility classes that effectively "replace" CSS classes by moving the equivalent CSS into your HTML. However, what they're not using is Tailwind's `@apply` directive [1] which basically encapsulates these repeated declarations with a repeatable, named CSS class. Doing so would replace their h3 tags with something like 1. Pruning and CSS of h3.header { @app…

I am now educated. Thank you.

Re: GPT-4 is phenomenal at Code

#213

Post by 1attice »

Jensson wrote: post #208
1attice wrote: post #190

Earlier quoted context omitted.

It will absolutely have great credibility. Did you read any of the code samples in the GH repo linked by OP? Here's the algorithm to get you appropriately scared. - Measure the amount of time it took to go from GPT 2 to GPT 3 - Measure the amount of time it took to go from GPT 3 to GPT 4 - Measure the code quality improvement for the same spans. Go ahead. Pick a metric. You'll notice that, not only is GPT-4 really fs…

> - Measure the amount of time it took to go from GPT 2 to GPT 3 1 year. > - Measure the amount of time it took to go from GPT 3 to GPT 4 3 years. > it represents a year-over-year improvement that is accelerating. I don't see how the above shows it is accelerating.

It does show, however, that one needs to read carefully.

Please evaluate the code quality metric, as I originally asked. (Your pick.)

Go look at GPT2's code.

Re: GPT-4 is phenomenal at Code

#214

Post by moonchrome »

CapstanRoller wrote: post #168
ramesh31 wrote: post #150

Earlier quoted context omitted.

>Sure, you may have a temporary advantage by leveraging AI and effectively doing labor arbitrage (the same way some craftsmen could gain a temporary advantage over others if they had access to rare/expensive tooling in the early days of the Industrial Revolution) but this advantage will quickly disappear as AI tooling becomes cheap and ubiquitous. "You only have to get rich once."

Adoption is happening way too fast for your approach to work, IMHO. I can see it being viable if you worked at some consultancy that invented ChatGPT and kept it to themselves, but at this point it's too late to contain it.

What's the adoption rate of copilot ? It's been out for how long now ?

Re: GPT-4 is phenomenal at Code

#215

Post by simonw »

angrais wrote: post #48
simonw wrote: post #30

Earlier quoted context omitted.

This is one of the reasons I'm so excited about being able to run large language models on my own laptop. I've managed to get LLaMA to generate snippets of Python code already, and with some fine tuning I imagine it could become equivalent to Copilot. https://til.simonwillison.net/llms/llama-7b-m2#user-content-...

Although that is an outdated generated snippet as it doesn't use pathlib? What if you added a type annotation to filename as Path?

Yeah it's a pretty terrible completion - GPT4 does that MUCH better.

But GPT4 doesn't run on my laptop using ~4GB of RAM.

Re: GPT-4 is phenomenal at Code

#216

Post by CapstanRoller »

moonchrome wrote: post #214
CapstanRoller wrote: post #168

Earlier quoted context omitted.

Adoption is happening way too fast for your approach to work, IMHO. I can see it being viable if you worked at some consultancy that invented ChatGPT and kept it to themselves, but at this point it's too late to contain it.

What's the adoption rate of copilot ? It's been out for how long now ?

>What's the adoption rate of copilot?

Growing very rapidly[1] with a userbase of millions[2] already

[1] https://www.ciodive.com/news/github-copilot-microsoft-softwa...

[2] https://d3.harvard.edu/platform-digit/submission/github-copi...

Re: GPT-4 is phenomenal at Code

#217

Post by simonw »

stuff20230314 wrote: post #49
simonw wrote: post #19

Earlier quoted context omitted.

Yes, it can do both of those things. I've done a bunch of stuff with ChatGPT where I've pasted in code and asked it to make changes. The main limit there is the token limit - 8,000 I think for ChatGPT - so I'm very much looking forward to experimenting with the increased 32,000 limit when that becomes available. I write tests for my code using Copilot a bunch - sometimes by pasting in the code I want to test and star…

1. So, current and immediate future limits prevent this beyond relatively small projects? 2. But like, you still have to write tests though? If I need to think and check what the generated tests are doing, I'd rather write them myself to be honest. I want the time-consuming parts automated, not typing code. Maybe I am in minority. Surely generating 99% complete unit tests should be some sweet-spot for ChatGPT? I woul…

Absolutely. Using LLM code without reviewing it is an even worse idea than accepting a PR from a brand new apprentice-level programmer at your company without reviewing it.

Re: GPT-4 is phenomenal at Code

#219

Post by sinity »

dougmwne wrote: post #204
kamaal wrote: post #196

Earlier quoted context omitted.

>>I think it can actually do all those things reasonably well. I don't you get it. Or even AI gets it. Having pet slaves is a very different feeling. It is not about getting work done. It is about the feeling of power, control. To both create and destroy. GPT won't be a problem for the same reason why programming itself never automated programmers out of jobs. The current state of programming is already good enough t…

This is actually an excellent point. Also why I think we will eventually all be back in the open office, even though I enjoy working in my pajamas.

Possibly. See https://thezvi.substack.com/p/escape-velocity-from-bullshit-...

The scariest thing is that there are people who advocate for it. Because humans are dangerous, I guess, so it's better to preemptively enslave them.

Sample

> Social control in the sense of not wanting lots of unemployed and restless youths. Having a system where long term and steady work is required in order to "live a good life" implies control - you have to act right and follow the rules in order to keep a job, which is itself necessary in order to have enough food and other necessities.

> Those making this argument here I believe are also making an argument for an alternative where the productivity of society is more equally spread, without the need to make everyone work for it.

> I agree that it's a system of social control, but I don't think it's nefarious or bad. We really don't want to live in a society where 25-year-old men don't have meaningful work to do and roam the streets getting into trouble.

The argument also doesn't _really_ make sense - there's already socially accepted system of 'social control' which _directly_ keeps people following the rules. The law.

Also unclear why lack of work would cause young people to "roam the streets" instead of staying home and roaming the internets. As they're already increasingly doing in free time.

Re: GPT-4 is phenomenal at Code

#220

Post by sinity »

ChuckNorris89 wrote: post #25
recuter wrote: post #14

Remember the comments about artists and creatives when DALL-E came out? I remember.

I think my barber and my plumber have safer jobs from automation than average coders.

> I think my barber and my plumber have safer jobs from automation than average coders.

Possible, but their 'advantage' is unlikely to last more than a few years.

Post reply on HN