Ask HN: Have you lost enthusiasm for programming and learning after ChatGPT?

[ducktective]

Personally, fantastic performance of LLMs and Stable Diffusion has hurt my motivation for reading documentations of cool libraries or studying another programming language book cause it seems to me all of these "classical" methods are going to be redundant soon.

I mean, how far are we from interactively describing the problem in English and having an AI translate it directly to ASM instructions? Aren't LLMs, "transformers" in essence?

Yeah, ChatGPT spits out hallucinations and downright wrong answers for some technical questions but it's only GPT3. Ridiculous amounts of funds is getting redirected to this space. They're gonna improve these models, they always do [1] [2].

And where do we stand here? Would you learn a book on "Optimal Design of Vacuum Tubes" when Shockley introduced the transistor? Would you read the documentation of a steamship when Tesla was demonstrating his coil to the public?

[1]: https://miro.medium.com/v2/resize:fit:651/1*aksFYLAhO-I85DST29svXQ.png

[2]: https://arxiv.org/abs/2302.14045

[damascus]

It's actually sparked my joy of coding again. I'm a solo consultant by day and I do a lot of technical work obviously so coming home and coding in my free time isn't something I've been super interested in recently. With GPT3 Davinci that changed and I started playing with it in python. That lead me to want to create an app that I've wanted to exist in the world [1] but that meant learning React Native (or similar but I went with RN) and that lead into MongoDB and Firebase and all sorts of other tech I'd heard about but never really played with. Anyway, the app I made is free but location specific to Seattle. My python backend pulls data from a lot of local venues (using BeautifulSoup web scrapers where APIs aren't available) and then uses (now) GPT3.5-turbo to either re-write or in most cases write original descriptions of the events the backend found. ATM its totally free with no ads because its a fun learning project. Now that the framework is in place I've been spending time with prompt engineering to create a more cohesive response across a wide range of inputs. It's been really fun!

[1]: http://www.letsgofrolic.com

[jameshiew]

It has not, although it has greatly changed how I program and learn. We are currently entering a period where "centaurs"[1] - humans using AI - will be the most effective agents for many tasks, including programming. With that in mind, I'm enthusiastic to embrace AI as a daily tool, knowing it will let me do things faster and better - but I don't expect this period to last forever and so it has put a sense of urgency in me to execute rather than just ideate, which I did not have before the dawn of things like ChatGPT.

[1]: https://www.parc.com/blog/half-human-half-computer-meet-the-...

[jaredsohn → jameshiew]

Not sure why that article went with centaurs; I think cyborg is a more accurate term. Just asked ChatGPT for a term for half human / half AI and it suggested cyborg as well.

[boredemployee]

I honestly don't understand people classifying gpt models as toys. I've been using it for my day to day so my productivity has increased tremendously. I no longer need to memorize a bunch of useless syntax or read documentation for the trendy dataviz library for plotting a graph in python, just to name one example. I confess that this undermined my desire to learn programming in a deeper way, but that's how life is at the moment, unfortunately the corporate environment requires us to execute and then think about it, it's the game I'm playing.

[hodgesrm]

> Have you lost enthusiasm for programming and learning after ChatGPT?

Quite the contrary.

ChatGPT may or may not be useful, but it demonstrates that AI will be an intellectually stimulating topic for decades to come. One of my kids is in college doing a degree in math and is also very interested in linguistics. She's intrigued by ChatGPT and now contemplating getting deeper into computer science. Her curriculum is heavy on discrete math, which seems like exactly the right background to understand LLMs in particular and AI in general. If good things come of this tech it will be at least in part due to the curiosity of people like her. It's hard not to feel optimism.

[dragonwriter]

> Yeah, ChatGPT spits out hallucinations and downright wrong answers for some technical questions but it's only GPT3.

No, its GPT-3.5.

> Ridiculous amounts of funds is getting redirected to this space. They're gonna improve these models, they always do

And it almost always takes decades longer than people expect in the early hype.

Really, aside from seeing “prompt crafting” as an additional important tool in the programmer’s toolbox, and opening up a whole range of new possibilities incorporating new technologies into apps, I don’t think it changes things for programmers that much in the near term, and even in the lo g term, while I’m sure there will be big changes, I think we’ll look back at “LLMs will eliminate the need for specialized programmers” the same way as the same thing being said about 4GLs and associated low-code/no-code tools.

[gimili]

In my opinion what we are seeing is the age of "Assisted Everything" [1].

I am very certain that in 2 years' time no white-collar job will be performed anymore alone in front of a PC. You will always collaborate with an assistant.

Personally this makes me super excited. As a Satellite Engineer, I always wanted to have a Jarvis-like experience and was super disappointed, when I found that the hardware engineering world was only paper-pushing excel files. Finally the future is here and engineers can focus on solving problems instead of on tasks that can be automated.

[1] https://assistedeverything.substack.com/p/the-age-of-assiste...

[nostromo123]

I don't understand what you guys are working on if ChatGPT can do your jobs...

[ducktective → nostromo123]

> Shockley invented the first working transistors at Bell Labs, the point-contact transistor in 1947

> The Intel 4004 is a 4-bit central processing unit (CPU) released by Intel Corporation in 1971

[gabrielsroka → ducktective]

> The world's first transistor computer was built at the University of Manchester in November 1953

[weatherlite → nostromo123]

If/when it learns to reason about a whole codebase why wouldn't be able to do everyone's job? For now the main block is its lack of context. It can work/remember and answer questions about a couple of functions maybe, not a codebase.

[sergiotapia]

Not at all! If anything GPT has uplifted me and now I can focus on the meat of my problems and not remember what this specific incantation looks like in this specific framework/language.

It's been a force multiplier for me, across the tech stack. Database, backend, frontend, edge, infra - you name it.

It's a great product.

Engineers who don't integrate this into their workflows will probably be left behind as much as we left behind engineers who manually manage their memory. (hint: not too much but their scope of work dramatically shrunk to specific nitches)

[jsz0]

Not programming but looking at the trends in DevOps and similar roles I don't see much of a future for it when a minimally technically literate person will be able to run AI based tools to solve complex/unique problems in the near future. There's always been a lot of knowledge churn in this profession but now it's hard to be motivated to learn the latest trendy thing when it's almost certainly going to be hidden under numerous layers of abstractions soon.

[roland35]

Nope! ChatGPT doesn't know how to do anything except put words together that seem plausible.

It is pretty good at some types of things no doubt, but let's not get carried away by an LLM!

[foothebar]

Yeah, I really waited for a machine which steals my job by providing great code like:

   for p in bn.parents:
     if p not in bn.parents:
       take_over_the_world()
…

[the_only_law]

Nah I lost that years ago, I am thinking more on what the hell else I could do for a living though, and the realistic options don’t look appealing at all.

[altdataseller → the_only_law]

How old are you and why do you think you have lost your desire for learning?

[weatherlite]

I'm 10 years in the field. I wouldn't say I stopped learning but it certainly made me stop and evaluate things a bit more, and slowly and carefully think about a plan b. The thing is it is so difficult to know what the job market will be like 10 years from now its pretty much impossible to come up with a solid plan b.

[gopisuvanam]

Programming is more than just writing code.. its about figuring what to build, architecting the solution and putting together code to build that.. ChatGPT in fact helps in the last part. It is in fact allowing me to write code faster and hence giving me more time to focus on the first two things..

[ofalkaed]

If ChatGPT decreased my enthusiasm for anything it would be the internet but I am the sort of person who has read books on the design and construction of vacuum tubes and did so long after Shockley and his transistor, it is interesting stuff.

[schwartzworld]

ChatGPT and code is just the mechanical turk. It's a very impressive parlour trick, but it's not going to replace human ingenuity

[_p4lt]

I'm certainly sick of reading about ChatGPT.

[gardenhedge]

No. In fact it has done the opposite for me. It's like having my own personal assistant and makes me more productive.

[speedgoose]

ChatGPT wrote some simple x86 Linux assembly for me, and explained well how it worked. And it worked.

[joshxyz]

it helped me code better again. i dont ask it for exact code, instead i ask it for ideas or guidelines on how to do xyz.like high level explanations.

[groffee]

It's not 'hallucinations', it's garbage in, garbage out (GIGO)

ChatGPT is about as far removed from AGI as it's possible to get. I wouldn't worry about it at all, it's just another tool, albeit a toy one at this point.

[ducktective → groffee]

What about this "emergent" intelligence when size of the model is large enough? There are blogs about it being able to model problems internally or forming a mini NN of its own.

As I've understood, even the researchers in the field have no idea why this happens. It's completely empirical and a "discovered phenomenon" at this point.

[1]: https://www.jasonwei.net/blog/emergence

[2]: https://jacobbrazeal.wordpress.com/2022/09/23/gpt-3-can-find...

[Jensson → ducktective]

You have millions of tutorials that are about a few pages long, I'm pretty sure it got its programming skills from those. But we are already at model size where the model can keep an entire tutorial page in its context, there is no extra benefit here from expanding the context size, it would need similar tutorials but much much longer and expansive to get better than it is now, and there just aren't enough humans to write all of those.

[_2uwr]

> I mean, how far are we from interactively describing the problem in English and having an AI translate it directly to ASM instructions?

Not Assembler but someone has documented their todo app on youtube. https://www.youtube.com/watch?v=Pi-5_eid7VA&ab_channel=Joshu...

The youtube comments are interesting, and I'd say, see how far you get trying to build an app with it meeting all the requirements, thats the real proof in the pudding.