Hacker News Reader

Comment by lebean | Hacker News Reader

I'm sorry to hear that it makes you sad. Maybe it's not empowering for many programmers, but stuff like GPT has inspired many non-programmers to learn programming just to use it. Which is pretty empowering.

View on HN

being a "prompt engineer" is hardly programming

I agree. I think we'll see similar results to 1980s when you needed $10k of equipment to make movies/videos vs today when Youtube and cheap storage/cameras democratizing video into "content". It's going to have a mix of breakthroughs and junk.

I think it's definitely coming, with more recent advances in unsupervised learning and multimodal learning (along the lines of https://github.com/microsoft/unilm or https://github.com/Alpha-VLLM/LLaMA2-Accessory)

Used to just mean 'shows up on time' ;)

I do think the idea of 'prompt engineering' isn't 'programming' in the low-level sense, but it's close enough for some peoples' needs to qualify, and will be a useful skill. But I think it'll be more like "being good at searching google" was a few years back. There was a period where you could be very productive understanding a few things about searching (filtering/keyword stuff, mostly) but that 'skill' isn't as useful today as google continues to put less emphasis on keeping those tools useful.

Similarly, being good with Excel. That's extremely powerful for a lot of people in their day to day jobs. Is it 'programming' in the classical hacker-at-a-desktop sense? No, but allowing people to get value from the computers in a way that's under their control (broad definition, I know) does, imo, fall under a large banner of 'programming'.

I take your point, but when you learn to actually code VBA in Excel, you get a lot more functionality.

Yep - VBA - you're in 'programming' territory for sure at that point.

To be fair they mentioned GPT and not ChatGPT.

That said I don't believe using ChatGPT to hobble together some python code to call the GPT API would constitute "learning to program" any more than nailing two boards together makes you a capable carpenter.

Sure, but there are people who have picked up Python, play with openai APIs, then realized how easy it is the glue stuff together, then all hell breaks loose. Branched off into using LLMs locally, or deep learning with fastai, or inspired to learn web dev, or whatever else. They may not get a job or really know what they're doing in the mere months they've put in, but that hasn't been too different from what some code bootcamps have produced. But it sure is empowering to people who, up to this point, haven't had a reason to learn Python.

If someone non programmer cab build useful things using chatgpt or similar tools. Its not too bad to call them programmer.

"prompt engineer" == my generation's "good at googling"

I can say from the OPs side, ChatGPT scripts are very helpful. They aren't perfect, but taking one and cleaning it up is usually quite a bit faster than writing it from scratch.

There's probably something to this, and other comments about AI being a source of competition. My experience with generative AI has been that it's neat, the academic papers are always a fun read, but I'm mostly underwhelmed by the output. This will get better with time (Sora is incredible), but it's missing the point. A recent anecdote: I'm in a small startup community and I asked the group what AI tools folks are finding useful. Someone mentioned that copilot makes them 2-3x more productive; personally, my tests with codellama has made me about 0.5-0.75x as productive as normal, mostly because I notice the mistakes and scrutinize the output more - probably a similar experience to folks who have been programming for a while.

But it must feel amazing for that 2-3x guy who is just trying to get his company off the ground, and that's great!

Aside: I recently went down a rabbit hole exploring fast food training videos from the 80s and 90s. It makes you appreciate the engineering (culinary, mechanical, and industrial) that allows a company to make a consistent product at scale, only requiring interchangeable, unskilled labor. Did you know that McDonald's claims that 1 in every 8 Americans have worked at the chain? You can get a fairly pricey jacket celebrating this fact[1]!

Perhaps we're finally at that point with computing. We know the externalities that have resulted from McDonalds and similar chains and, good or bad, we've accepted them. Over the next decade we get to watch the same with commodity knowledge work.

[1] https://goldenarchesunlimited.com/products/1-in-8-alumni-jac...

It's fantastic at creating boiler plates to be used when learning something new. I no longer get dragged to some site with information hidden behind a paywall or have to waste time churning through irrelevant YouTube search results, which is pretty much all YouTube search results at this point.

Reply on news.ycombinator.com