🆕 blog! “GitHub's Copilot lies about its own documentation. So why would I trust it with my code?”

In the early part of the 20th Century, there was a fad for "Radium". The magical, radioactive substance that glowed in the dark. The market had decided that Radium was The Next Big Thing and tried to shove it into every product. There …

👀 Read more: https://shkspr.mobi/blog/2024/10/githubs-copilot-lies-about-its-own-documentation-so-why-would-i-trust-it-with-my-code/

#AI #github #LLM

GitHub's Copilot lies about its own documentation. So why would I trust it with my code?

In the early part of the 20th Century, there was a fad for "Radium". The magical, radioactive substance that glowed in the dark. The market had decided that Radium was The Next Big Thing and tried to shove it into every product. There were radioactive toys, radioactive medicines, radioactive chocolate bars, and a hundred other […]

Terence Eden’s Blog
@Edent "My hammer has trouble driving this screw! So why would I trust it with this nail?"
@louis I'm curious, did you read the blog post, or just the headline?

@Edent I was replying to the mastodon post, not the blog post.

However, I just read the blog post, and I stand by my comment. You claim that trying to extract accurate information from an LLM is a basic functionality test, but LLMs aren't fact/truth machines and never have been and never will be (the screw in my analogy). But they are pattern matchers and that is immensely useful as a coding tool (the nail).

@louis next time, please read the article before replying.

I'll mute you for a bit. But feel free to unfollow me.

@Edent I experience something like this way too often, when I try some hyped AI tool: Can't google some info? Some AI-powered search engine provides it instantly — only for me to realise that it's hallucinated and the provided source does not contain the alleged info. Need to rephrase a text for simpler language? Some text-rewriting AI-tool does that and cuts out 75% of the information. And yet people seem happy with that search engine and that writing tool. Maybe I'm too dumb 🤷
@Edent "this mission is too important for me to allow you to jeopardize it."
@Edent This is the perfect parallel analogy. Damn.
@wcbdata I wish I were clever enough to have come up with it!

@Edent It’s odd that it’s so awful for some things, but can be really useful for some other things.

I wanted to update a .net3 project to .net6. I asked it how to do it, and then each time I got an error message asked it how to correct that error message and I ended up with the conversion successfully done in tens of minutes rather than days.

Would I let it loose on writing something from scratch? No way! But for simple-ish tasks that I’m unfamiliar with, it has been useful to me.

@Edent

In terms of the bigger picture analogy you draw with Radium though - I hate that it is being crowbarred into nearly every software package under the sun. Recently Canva has announced a pricing increase to pay for their ‘AI’ stuff (whether you use it or not, it seems) https://www.theverge.com/2024/9/3/24234698/canva-price-increase-300-percent-ai-features

Canva says its AI features are worth the 300 percent price increase

The price of team subscriptions are set to skyrocket next year, which Canva justified due to the apparent value that generative AI tools have added to the platform.

The Verge
@Edent I had this happen to me when I was using Coursera’s bot to try and work on gaps in my understanding before I took a quiz. I asked for some links to do additional reading on a subject. It gave me three, and all three were completely made up.