Charcoal HQ

« first day (3490 days earlier) ← previous day next day → last day (337 days later) »

12:49 AM

Enthiran (transl. Robot) is a 2010 Indian Tamil-language science fiction action film written and directed by Shankar. It is the first instalment in the Enthiran film series. The film stars Rajinikanth and Aishwarya Rai Bachchan with Danny Denzongpa, Santhanam and Karunas in supporting roles. The soundtrack album and background score were composed by A. R. Rahman while the dialogues, cinematography, editing and art direction were handled by Madhan Karky, R. Rathnavelu, Anthony and Sabu Cyril respectively. The story revolves around the struggle of a scientist K. Vaseegaran to control his android...

@Undo Can you please edit the chat bios of SmokeDetector and metasmoke on all three chat servers so they reflect the strike?

Might also want to edit the site bios.

3 hours later…

Cody Gray - on strike

3:45 AM

@RyanM-Regenerateresponse First coffee, now tea. I see what everyone means about generative AIs being inconsistent and unreliable.

user16217248

4:16 AM

So. Much. AskUbuntu. Spam.

Cody Gray - on strike

4:29 AM

Oh, excellent news.

user16217248

What is it? Did the staff concede that their AI policy was bad?

2 hours later…

Andrew T.

6:34 AM

I thought "excellent news" meant that "the strike was working"?

2 hours later…

Cody Gray - on strike

8:23 AM

Yes, that's what it meant.

8 hours later…

user16217248

4:42 PM

It would be better for the strike to "have worked" than the strike to "be working"

I.e. for the AI policy to be repealed by the staff and business to go back to normal

VLAZ -on strike-

5:04 PM

Even better, if there was no strike. But it is what it is...

4 hours later…

Cody Gray - on strike

9:19 PM

Yes, I agree, but my time machine does not seem to be working, and so here we are. I'm open for suggestions.

Laurel

9:47 PM

@CodyGray-onstrike You should know better than anyone that "it doesn't seem to be working" is not acceptable here. Your question is too broad. Please provide a minimal example of your code.

1 hour later…

Johannes Kuhn

10:57 PM

https://meta.stackexchange.com/questions/389928/gpt-on-the-platform-data-actions-and-outcomes
Personal opinion: It all hinges on the "we have a statistical model that doesn't match the actions taken".
Problem: SO was quick to point out that obvious ChatGPT answers were deleted and users suspended. Some users may work around it by modifying the chatgpt response (possibly on StackOverflow).

Ryan M - Regenerate response

In particular, it hinges on that statistical model accurately predicting the absolute magnitude of AI-generated posting.

especially doing so in the face of major and likely unrelated changes in behavior that their theory does not explain, such as a drop in questions

Johannes Kuhn

Yes. Exactly.
Interpreting statistical models is *very* hard, and it is easy to come to the wrong conclusion.
They did come to a conclusion, but I think they came to the wrong conclusion.

If there is such a high discrepancy between generated answers and moderator actions, then picking some random suspension for review should confirm this.

But "because ChatGPT detectors are unreliable" and even "human ChatGPT detectors are unreliable", it is "impossible" to figure out if a post has been written by ChatGPT.

And now the big problem: Unless they can admit that they were wrong (we are talking about SE here), I don't see how that can be solved at all.

Ryan M - Regenerate response

11:16 PM

meanwhile actual ChatGPT posters are just slinging out recommendations to use methods that don't exist left and right

Admittedly humans do that too, but it's usually, like, a typo or a slight misremembering of the name of a real method

not like, a wholly hallucinated API

but sure, maybe they're sleepwalking, had a dream where the API existed, and decided to answer a Stack Overflow question.

We cannot, with certainty, rule out the possibility that the answers were written by the Ambien Walrus and not an LLM.

Laurel

@RyanM-Regenerateresponse The spite in me wants to recommend you escalate those "suspicions and hunches" to staff, but they would probably just quickly tell you to not

Ryan M - Regenerate response

It would be helpful to get a list of confirmed, obvious stuff to show

Particularly in higher volume than they claim the true numbers are

I've found one user who has posted 14 obviously ChatGPT answers in the last hour. So that pretty much accounts for your entire daily quota even if no-one else posts anything. — DavidW yesterday

Johannes Kuhn

I have seen an answer that invented "a java method is marked accessible".

How did they validate their statistical model when they can't distinguish between ChatGPT answers and real answers?

Ryan M - Regenerate response

As far as I can tell, the only validation of that model is that the metric spiked with the release of ChatGPT.

Cody Gray - on strike

11:34 PM

And inflection point was exactly on Dec 8...

« first day (3490 days earlier) ← previous day next day → last day (337 days later) »

all rooms