Simon Willison interview: AI software still needs the human touch Simon Willison, a veteran open source developer who co-created the Django framework and built the more recent Datasette tool, has become one of the more influential observers of AI software recently. His writing and public speaking about the utility and problems of large language models has attracted a wide audience thanks to …
"There's the moral, ethical side and there's a legal side and they're not necessarily the same thing, you know. Things could be legal and still feel wrong."
Do you really think that will stop -- has ever stopped -- a corporation when it's standing in the way of profits? "Don't be evil" ...
"I have a hunch that programmers, software engineers, are the group best served by this technology right now. We get the most benefit from it. And part of the reason for that is that these things are notorious for hallucinating. They'll just make stuff up. If they hallucinate code, and you run the code, and it doesn't work, then you've kind of fact checked it."
I think the real problems might arise if the LLM has hallucinated code and it does run. Now you've got a thornier problem on several fronts: does it produce correct results in all cases; does it handle edge-case inputs correctly; does it behave appropriately when an error is thrown; etc.
The hardest bugs to find seem to be the ones where the code compiles and runs -- but not always correctly. And now you're picking through someone (or something) else's code, trying to fully understand the underlying logic ...
I think it's just that the interviewee is a software person, giving him a biased viewpoint. Also, if you waited until running the code to find out that Copilot had made a mistake, you would be wasting time more time than necessary.
Many writers (of all kinds) are using it successfully, so long as they keep themselves in the loop, otherwise they will lose their employment or contracts.
It's not new - Michelangelo had several hired assistants helping him in the Sistine Chapel. Of course he made sure he was directing the artistic content. Just to be clear, I personally shut my eyes and curse every time I see another AI picture used by El Reg. I love human art only and refuse to look at anything else.
The one thing that might be making coding application of AI move faster is that AI coders tend to improve their own tools first. That was true before AI came along as well.
@Grogan
"If it does somehow work, and you can't follow it or understand it, how are you going to maintain it?"
That could be applied to lots of code where something has been written by other developer(s), and when you discover it, you look at it and just scratch your head... I'm sure, before the days of code reviews, some people deliberately wrote obfuscated code as a job preservation tactic.
Using automation to get rid of tedious tasks is a godsend. Be it Ghat-GPT or other tools. For example, El Reg supports formatting with HTML, but what kind of madman would sit there and hand-code HTML for a one-time post? Much better to have an LLM do it for you.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
