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

31–40 of 107 posts

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

#31

Post by jkukul »

CapsAdmin wrote: post #21

On a slightly related note, does anyone feel copilot is not as powerful as it used to be? It's kind of hard to pinpoint exactly what it is, but it feels like it often either does nothing, or generates the bare minimum. I think one concrete thing is writing a comment about a function, then expecting that function below the comment, but instead you get more comments.

Agree 100%. I noticed that on the Copilot settings page [1] you can switch to Claude Sonnet model (instead of a model trained by Github I assume?). In my experience this improves things.

[1] https://github.com/settings/copilot

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

#33

Post by conartist6 »

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.

It's to draw attention away from the fact that LLMs are amoral.

You clearly have a moral spine because you're standing up and asking questions about what is right.

LLMs can't do anything like that so they need to be constrained by more direct methods in order to not appear to be evil

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

#34

Post by tristor »

RicDan wrote: post #18
1oooqooq wrote: post #15

Earlier quoted context omitted.

yeah, for sure my friend. the populistic election fear of "the gays" wich led to the election of a guy who bribed pornstars and another not elected but who yourself already recognize as the true ruler, who raised a transgender daugther, they are alrady omnipresent as gods and impacting every aspect of everyone's life where a pronom may appear. sigh.

You are seriously underselling the potential impact of things happening has on planning. Any company worth their 2 cents has contingency plans ready for several situations - i.e. if President A wins vs B, if there is enough plausibility that it might come with serious impact to the business.

> Any company worth their 2 cents has contingency plans ready for several situations

My guy, most major companies struggle to get a quarterly plan in place for what they're going to do by the start of the following quarter. There may be some department focused on lobbying and political risk that cares about this stuff, but there are no product and engineering teams at any company that have "contingency plans" just laying around to be activated for random shit like this. Nobody has time for that.

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

#35

Post by nightfly »

whywhywhywhy wrote: post #29
CapsAdmin wrote: post #21

On a slightly related note, does anyone feel copilot is not as powerful as it used to be? It's kind of hard to pinpoint exactly what it is, but it feels like it often either does nothing, or generates the bare minimum. I think one concrete thing is writing a comment about a function, then expecting that function below the comment, but instead you get more comments.

Seems the trend with most LLM tech is once the users are there the model is quantized or downgraded down to the bare minimum level of usefulness either silently or through new versions that are just not better in real use.

See also: search engines, streaming platforms, etc...

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

#36

Post by emsign »

CapsAdmin wrote: post #21

On a slightly related note, does anyone feel copilot is not as powerful as it used to be? It's kind of hard to pinpoint exactly what it is, but it feels like it often either does nothing, or generates the bare minimum. I think one concrete thing is writing a comment about a function, then expecting that function below the comment, but instead you get more comments.

I didn't notice any decline, it still gives me code that is wrong almost every time.

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

#37

Post by bshacklett »

whywhywhywhy wrote: post #29
CapsAdmin wrote: post #21

On a slightly related note, does anyone feel copilot is not as powerful as it used to be? It's kind of hard to pinpoint exactly what it is, but it feels like it often either does nothing, or generates the bare minimum. I think one concrete thing is writing a comment about a function, then expecting that function below the comment, but instead you get more comments.

Seems the trend with most LLM tech is once the users are there the model is quantized or downgraded down to the bare minimum level of usefulness either silently or through new versions that are just not better in real use.

Just a data point my experience with ChatGPT has only gotten better and better.

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

#38

Post by cheschire »

tristor wrote: post #34
RicDan wrote: post #18

Earlier quoted context omitted.

You are seriously underselling the potential impact of things happening has on planning. Any company worth their 2 cents has contingency plans ready for several situations - i.e. if President A wins vs B, if there is enough plausibility that it might come with serious impact to the business.

> Any company worth their 2 cents has contingency plans ready for several situations My guy, most major companies struggle to get a quarterly plan in place for what they're going to do by the start of the following quarter. There may be some department focused on lobbying and political risk that cares about this stuff, but there are no product and engineering teams at any company that have "contingency plans" just la…

Indeed, because good contingency planning inherently means wasting effort on if/else paths that may never occur.

I haven’t seen any recent OKR or KPI that was able to justify or even quantify the value of a path in history that didn’t occur.

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

#39

Post by pjc50 »

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.

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 their behalf.

Not to mention questions like "who is liable if an LLM libels someone".

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

#40

Post by sfn42 »

CapsAdmin wrote: post #21

On a slightly related note, does anyone feel copilot is not as powerful as it used to be? It's kind of hard to pinpoint exactly what it is, but it feels like it often either does nothing, or generates the bare minimum. I think one concrete thing is writing a comment about a function, then expecting that function below the comment, but instead you get more comments.

I've only tried copilot once and this is exactly my experience. I'd write a comment to try to prompt it, then it would keep writing the comment instead of the function. Best I could do was make it write the function as a comment then uncomment it.

I prefer to just write code myself

Post reply on HN