Using AI for the First Draft

Question

Using AI for the First Draft

Steve Jones - SSC Editor

SSC Guru

Points: 735195
More actions
June 5, 2024 at 12:00 am

#4423851

Comments posted to this topic are about the item Using AI for the First Draft

Viewing 4 posts - 1 through 3 (of 3 total)

You must be logged in to reply to this topic. Login to reply

David.Poole SSC Guru Points: 75939 More actions · Answer 1

With a few exceptions, I've found the time, gained by generated code, is lost by the manual process of debugging and refactoring.

Using LLM for Java/C# etc strikes me as trying to use LLM questions as a declarative language to abstract the complexity of the underlying language. Like SQL does for interacting with RDBMS .

LinkedIn Profile

Rod at work SSC-Dedicated Points: 33939 More actions · Answer 2

I've used generative AI, such as GitHub Copilot, at home, for analyzing code. Both to tell me what a method is doing and where the problems in the code are. I find that helpful, but it doesn't go the whole 9 yards. However, its good enough so that I'll continue to use gen AI at home.

Unfortunately, my employer won't pay for it, so at least for the next few years we won't be able to benefit from using it.

Kindest Regards, Rod Connect with me on LinkedIn.

Jeff Moden SSC Guru Points: 1004052 More actions · Answer 3

I agree with David Poole. To put it all in less formal terminology, I've found that tools like ChatGPT are nothing more than a very sophisticated consensus engine and, sadly the source of that consensus is the same source if you do a normal search engine search. The consensus frequently overwhelms the correct outlier sources... and some of the answers are actually dangerous in one form or another.

I've been checking ChatGPT 3.5, ChatGPT 4o, CoPilot, Gemini, and a few others. They all very confidently fail by producing code that will "Silently Fail" (produces incorrect answers in some instances with no indication that an error has been made) to the question of "In SQL Server (T-SQL), what is the code to convert a DATETIME2 value to DATETIME". They ALL got it wrong. Some do correctly point out that resolution may be a factor but the code they produce will provide the incorrect answer in many cases that most people (even "experts" on the Internet) never test for.

Of course, it's not just that one question.

The really interesting part is how people justify its use to augment coding. The problem is that a lot of programmers that actually don't know how to do something, won't know that the very confident sounding answers of the monster before them is, in fact, wrong.

Of course, that would also be true of people using normal search engines. AI is just a quicker and more expensive manifestation of the same problems. Ironically, that makes it "more human". 😉

--Jeff Moden

RBAR is pronounced "ree-bar" and is a "Modenism" for Row-By-Agonizing-Row.
First step towards the paradigm shift of writing Set Based code:
________Stop thinking about what you want to do to a ROW... think, instead, of what you want to do to a COLUMN.

Change is inevitable... Change for the better is not.

Helpful Links:
How to post code problems
How to Post Performance Problems
Create a Tally Function (fnTally)