92% of programmers are using AI tools, says GitHub developer survey

[Flash]

https://www.zdnet.com/article/github-de ... -ai-tools/

[bigpup]

MY AI gave you a thank you!

[amethyst]

Tried Reverso, an AI language translator to translate a book. Still need lots of work (or shall we say "intelligence"). Poor in comparison to Google Translate for example. Relatively strange since this is one thing that one would have thought it would be very accurate in.

[dimkr]

Interesting, a company that sells x says that 92% of its target audience uses x.

I wonder if the definition of "AI tools" includes things like GitHub's search bar and not just Copilot, and I wonder if data from "500 U.S.-based developers at companies with 1,000-plus employees" means mostly data from junior developers, who spend most of their time on new code and not architecture, refactoring, performance optimization and technical debt. IMO programming is easy, but true software engineering is another level, and I still haven't seen AI that truly helps with the latter.

[rockedge]

IMO programming is easy

Many people would tell me my job setting light for cinematography was just turning a switch on and off.

[dimkr]

IMO programming is easy

Many people would tell me my job setting light for cinematography was just turning a switch on and off.

Many people tell me my job can be automated, because they think I only write code

[sonny]

"The craft of storytelling is the most complex craft in the world."
For AI? No way.

[mikeslr]

Following their leaders, periodically 92%* of lemmings run off cliffs to their deaths. It's nature's way of preventing that species from exceeding the annual biomass created and necessary to support the healthy continuation of that species.

Compare that with invasive rats on islands: first they eat everything else; then they kill and eat each other; then they go extinct.

“those whom God wishes to destroy he first drives mad” Boswell, perhaps paraphrasing a Roman adage.

* Poetic license was used as to the amount. Nothing else.

[pp4mnklinux]

Test this and then comment if you want. https://www.useblackbox.io/chat

I don't know how useful could it be... but I suppose it is a good resource.

[mikewalsh]

I've also been using AI tools in my coding, and they help. They make coding faster and easier, especially when stuck on a challenging problem. It's astonishing how they better understand what I want to do with my code than older tools.

The only problem with doing that is that you never truly learn how to cope with any given issue in future. Slowly but surely, you become lazy.....and gradually surrender your creativity to a machine. It's not really the case that they "understand better"; they're just better - and a lot faster - at using all the available search engines than you or I are. They re-hash input, and try to make it sound like an intelligent response. Much of the time they're just quoting your own input back at you....but so many folks see that as intelligence, because the machine is interacting with you on your own level.

I don't see what all the fuss is about. What price real "creativity", hmm?

Mike.

[dimkr]

Most people are bad or at least not excellent in their line of work (empirically and according to the law of large numbers). If you're good at what you do, there are probably millions who do it much better.

The quality of code is not some a-priori truth, it's always tied to time and place. Some old and "bad" languages or frameworks can be very productive for certain kinds of development - some organizations still have internal tools written by contractors who use Visual Basic. And security-related programmer sins can be OK in code that runs in a closed, very deterministic environment. Now add the dimension of time: the "latest and greatest" can be a good fit for one project (for example, a startup company that wants to sell now and doesn't care about the future), while "boring is good" tech can be good for another (for example, software for the financial sector), and some languages (like C++ or Java) evolve over time, so different codebases can be wildly different in both style and runtime behavior.

The corpus this AI learns from consists of code written by people in various places, for various purposes, over many years, and it doesn't know all this context. It doesn't know if security, efficiency or long-term maintainability of the code was a priority, and some programmers simply don't care about these things and write code full of footguns.

Do the 1+1 and you'll reach the conclusion that the auto-completed code written by AI is extremely unlike to be state-of-the-art, and can be super bad or even dangerous.

[rockedge]

@dimkr Well said. Makes total sense.