Skip to main content
296 votes

GPT on the platform: Data, actions, and outcomes

What makes you so confident that your methods of identifying GPT posts are more accurate than how moderators have been identifying them? So far, all of the data I've seen hinges on the assumption that ...
Mithical's user avatar
  • 84.5k
262 votes

GPT on the platform: Data, actions, and outcomes

Thank you for posting the methodology! That's a step forward. Unfortunately, there's a critical flaw in your methodology: you assume that In principle, if people are copying and pasting answers out ...
Gilles 'SO- stop being evil''s user avatar
231 votes

GPT on the platform: Data, actions, and outcomes

Appeals of moderator actions that cannot be validated need to be shown to the moderator team When taking action against a user, we need to have strong evidence that we are correct. That evidence ...
Ryan M's user avatar
  • 28.6k
201 votes

GPT on the platform: Data, actions, and outcomes

Notes about my following SEDE queries: Upon new SEDE refresh, you may have to run the cross-site queries a few times for them to finally complete). Ignore the datapoints for the current month (cropped ...
starball's user avatar
  • 29.2k
143 votes

GPT on the platform: Data, actions, and outcomes

I came off of read-only mode for this. My initial take on this is that the analysis tries to affirm that user retention is valued higher than user quality, but the data points are still reasonable ...
Makoto's user avatar
  • 58.3k
126 votes

GPT on the platform: Data, actions, and outcomes

Copying my response from Teams, with some small edits to remove things that you chose not to reveal publicly for some reason: I am skeptical about several points in your analysis. But most troubling ...
Chris's user avatar
  • 2,994
122 votes

GPT on the platform: Data, actions, and outcomes

What percentage of users who have been suspended (for any amount of time) for GPT use were active answerers as you described them immediately prior to ChatGPT's introduction? I wasn't a very heavy ...
Kevin B's user avatar
  • 11.9k
87 votes

GPT on the platform: Data, actions, and outcomes

What about everything else? We've known for a long time that detectors (HuggingFace is the one I'm most familiar with, but others as well), have an extremely high rate of false-positives. Having data ...
cocomac's user avatar
  • 16.2k
85 votes

GPT on the platform: Data, actions, and outcomes

The "actual GPT posts have fallen" graph is the lynchpin of your entire analysis. Without that graph, you've proven exactly nothing (aside from "engagement is down and the CM team has ...
Kevin's user avatar
  • 5,907
84 votes

GPT on the platform: Data, actions, and outcomes

OK, finally there is something to actually discuss. This discussion should have happened before the policy was decided, and the policy should have been decided with community-elected moderators' input ...
kaya3's user avatar
  • 6,286
74 votes

GPT on the platform: Data, actions, and outcomes

We are, as of right now, operating under the evidence-backed assumption that humans cannot identify the difference between GPT and non-GPT posts with sufficient operational accuracy to apply these ...
NotTheDr01ds's user avatar
  • 4,980
72 votes

GPT on the platform: Data, actions, and outcomes

My summary of your nine-google-doc-page question You have tested how the ChatGPT-detectors work on human written posts and you did not like the accuracy. You came up with a heuristic that you use to ...
Nicolas Chabanovsky's user avatar
71 votes

GPT on the platform: Data, actions, and outcomes

I already posted an answer to this question, in which I try to assert why people create posts using GPT and why others do not find that behavior healthy for the community. I still wanted to post ...
CodeCaster's user avatar
  • 4,931
69 votes

GPT on the platform: Data, actions, and outcomes

The one-size-fits-all policy is a problem The data you present is somewhat compelling. An answer with many cached versions, showing the answer being fleshed out over time, is a pretty reasonable ...
cag51's user avatar
  • 1,341
64 votes

GPT on the platform: Data, actions, and outcomes

Thanks for elaborating on your reasoning. However, I still have questions. How was urgently overriding the moderators' mandate a conclusion? Your exposition does not at all reveal how such an ...
tripleee's user avatar
  • 15.1k
60 votes

GPT on the platform: Data, actions, and outcomes

What follows is the internal ‘gold standard’ for how we measure GPT posts on the platform. While I'm still digesting much of the data, you seem to be failing to consider alternative explanations for ...
NotTheDr01ds's user avatar
  • 4,980
59 votes

GPT on the platform: Data, actions, and outcomes

there is no future for the network at all. You might be closer to the truth than you think, or fear. Regardless of the above, no Community Manager will tell you that removing 7% of the users who try ...
CodeCaster's user avatar
  • 4,931
58 votes

GPT on the platform: Data, actions, and outcomes

This is based on a question that I asked on the Stack Moderators Team. It wasn't addressed there, but it may have gotten lost in the shuffle. I've trimmed it down to make it more appropriate for Meta ...
Thomas Owens's user avatar
  • 55.4k
51 votes

GPT on the platform: Data, actions, and outcomes

I have big doubts about this part of presented data: Yet, at the same time, actual GPT posts on the site have fallen continuously since release It's implied that posts with a small number of Drafts ...
markalex's user avatar
  • 1,791
48 votes

GPT on the platform: Data, actions, and outcomes

Analogy and motivation Let's say I'm a dairy farmer in Wisconsin. I look at some data about milk output from my dairy and find that output is down 50% from last year. This is a remarkable drop, and it ...
Bryan Krause's user avatar
  • 14.6k
47 votes

GPT on the platform: Data, actions, and outcomes

TL;DR: Your analysis that concludes “automated GPT detectors have unusable error rates on the platform” is not sound. I can’t easily find what exactly Hugging Face’s percentage signifies, but I assume ...
Wrzlprmft's user avatar
  • 28.3k
45 votes

GPT on the platform: Data, actions, and outcomes

The methodology presented is very interesting, but if the company really believes that human beings can't tell the difference between good quality posts that contribute to the site and poor quality or ...
ColleenV's user avatar
  • 27.1k
44 votes

GPT on the platform: Data, actions, and outcomes

since the advent of GPT, the % of content produced by frequent answerers has started to collapse unexpectedly The data you show, along with system changes introduced since advent of GPT suggest that ...
gnat's user avatar
  • 11.3k
44 votes

GPT on the platform: Data, actions, and outcomes

There are multiple things I have questions/concerns about, but I'd like to focus on just one. This still seems like throwing the baby out with the bathwater with regard to what the announcement/policy ...
TylerH's user avatar
  • 16.9k
37 votes

GPT on the platform: Data, actions, and outcomes

Thank you for providing some details, but I have several issues, mainly statistically related. First, there's with In order for us to consider using detectors of any kind (automated or human) on the ...
John Omielan's user avatar
  • 14.9k
37 votes

GPT on the platform: Data, actions, and outcomes

In the Community Management industry, it is a well-known fact that removing a person from a community, even for a short time, has an outsize impact on the contributor community. In other words - ...
mason's user avatar
  • 1,423
36 votes

GPT on the platform: Data, actions, and outcomes

It is worth noting that, early in the release of GPT, we changed the Stack Overflow rules to require new users to wait 30 minutes between first posts, instead of 3 minutes as was originally set for ...
Henry Ecker's user avatar
  • 5,000
36 votes

GPT on the platform: Data, actions, and outcomes

I've read this twice, pondered for a while on a few of your more vague turns of phrase. Most of my specific constructive thoughts and criticisms have already been covered by other users. I found Kaya3'...
BryKKan's user avatar
  • 1,281
34 votes

GPT on the platform: Data, actions, and outcomes

As to the future of the site part, I think it's imperative to be able to detect AI generated answers. Whilst it is true that voting can take care of hallucinations and plain incorrect answers, the ...
Adriaan's user avatar
  • 1,658
33 votes

GPT on the platform: Data, actions, and outcomes

This is basically the same thing Mithical already said, and it is the first thing that came to mind as soon as I started to read your very long post. That said, I am posting this version to give a ...
ꓢPArcheon's user avatar
  • 34.8k

Only top scored, non community-wiki answers of a minimum length are eligible