Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Copilot stops working on code that contains hardcoded banned words from GitHub (2023)

github.com

Post reply on HN
Print view View original

71–80 of 107 posts

Re: Copilot stops working on code that contains hardcoded banned words from GitHub (2023)

#71

Post by belter »

alkonaut wrote: post #68

Seems to work again. But how does this happen in the first place. How could someone possibly have thought "hey I have an idea, let's put in a list of english words and just silently stop working if we have see even one of them in a substring". And people in this meeting would nod and say "yeah that sounds like an easy safety fix, let's do that". This just feels odd. This isn't a piece of forum software written by a 1…

Because it was a Manager idea and the person responsible for implementing was a H1B ?

Re: Copilot stops working on code that contains hardcoded banned words from GitHub (2023)

#72

Post by lolc »

pocketarc wrote: post #57

One of the companies I worked with developed software for drug rehabs. Copilot would just constantly stop autocorrecting whenever I went into code files that mentioned anything to do with drugs (or sex - that's not a new censor!). It's the main reason I switched to Supermaven (before they got bought out and gave up on updating their extension).

What do you use now? I'm bummed by Supermaven not working well anymore but haven't looked around again yet.

Re: Copilot stops working on code that contains hardcoded banned words from GitHub (2023)

#73

Post by VonGallifrey »

StableAlkyne wrote: post #41
mycall wrote: post #17

I have a simple question. If censorship is considered evil regarding the written word and communications between humans, why do we want to then censor LLMs differently? It is either counterintuitive or simply a false concept we should abandon. Perhaps it is more about training, similar to how children are 'monsters' and need to be socialized/tamed.

> why do we want to then censor LLMs differently? Because of the obvious PR implications of having a program one's company wrote spewing controversial takes. That's what it boils down to - and it's entirely reasonable. Personally, I wish these things could have a configurable censorship setting. Everyone has different things that get under their skin, after all (and this would satisfy both the pro-censor and pro-unce…

> We both guessed the censor was catching it as a false positive for the slur.

There is a word for this. It is called the Scunthorpe problem. Named after the incident in which the residents of the Town Scunthorpe could not register for an AOL account because AOL had an obscenity filter that did not allow the Town name.

It has been a problem since 1996 and still causes problems.

Re: Copilot stops working on code that contains hardcoded banned words from GitHub (2023)

#75

Post by hbossy »

xyproto wrote: post #14

I wonder if one can sneak in words into code to avoid public code to be used for training AI. Perhaps just a tiny little nazi reference.

There's a +n....r license. It mandates that all copies of code include the n-word, so no big corporation can use it in their products.

Re: Copilot stops working on code that contains hardcoded banned words from GitHub (2023)

#76

Post by nonethewiser »

I can see why model providers wouldnt want to generate certain text for their own interests.

But lets not pretend its for the benefit of users. If a company could release an unmoderated model without real risk to themselves then they should do so.

Re: Copilot stops working on code that contains hardcoded banned words from GitHub (2023)

#78

Post by ben_w »

sfn42 wrote: post #49
pjc50 wrote: post #39

Earlier quoted context omitted.

They're not being censored by the state in this case (unlike the gotcha everyone keeps using when deepseek is mentioned). They're being limited by their developers for brand safety. There's still career-ending possibilities from saying things even in the overtly antiwoke government: https://www.independent.co.uk/news/world/americas/us-politic... ; people are not going to trust an LLM which might say such things on th…

That's easy. The person who used an LLM to generate the text and then published it as their own without even proofreading it is liable for the text they published.

Are you sure it's "easy", rather than "varies by jurisdiction"?

https://www.theguardian.com/technology/2024/nov/04/google-me...

Re: Copilot stops working on code that contains hardcoded banned words from GitHub (2023)

#79

Post by gedy »

alkonaut wrote: post #68

Seems to work again. But how does this happen in the first place. How could someone possibly have thought "hey I have an idea, let's put in a list of english words and just silently stop working if we have see even one of them in a substring". And people in this meeting would nod and say "yeah that sounds like an easy safety fix, let's do that". This just feels odd. This isn't a piece of forum software written by a 1…

> But how does this happen in the first place

Companies that are ironically filled with privileged people who "Gotta do something", since they, while sort-of well meaning, are sheltered and disconnected from actual social struggles in real life.

Post reply on HN