61

This post is (finally) to announce that I have formally rescinded the policy on Generative AI (specifically the one regarding GPT content detectors, which was announced on the moderators team and recently released here, under the title "(Historical) Policy on the use of GPT detectors"). That policy was obviously flawed in a number of ways, and I'm pleased to finally be able to vacate it in favor of this interim policy.

I would also like to take this opportunity to extend my most sincere personal apologies to mods who felt that in our previous text we were accusing them of racism. While that was not the intent of the text that I wrote (nor did that sentiment reflect the feelings of anyone involved in drafting the text), I can understand how it could be read that way, and I regret that we allowed it to be published like that. You have my sincere apologies, which I will also deliver in person at the upcoming mod/staff meetup.


The policy on AI-content detection reports

The intent is that the prior policy is vacated in favor of this interim policy, which will stand only until the working group of mods, staff, and other interested parties have finished their work and emerged with a new policy recommendation, which will take effect as the permanent policy at that time.

Until the permanent policy is adopted, the Community Management (CM) team will be accepting the following heuristics for reports and/or appeals of AI-crafted content. We will accept any combination of ONE of the "strong heuristics" OR several of the "weak heuristics". Please note that there are a few things that we will not accept at all, and moderators have been informed as to what those are.

Moderators are not expected to submit these heuristics for each action they take, however, it is expected that if the CM team reaches out to contact a moderator to discuss an action, the moderator will be able to provide the heuristics used to make the determination at that time, so that the CM can effectively double-check and/or explain the action.

Option one - the report/appeal/response includes at least one of the following currently agreed strong heuristics:

(The heuristics are not being released here, at the request of the negotiators for the Moderators.)

Option two - the report/appeal/response includes several of the following weak heuristics:

(The heuristics are not being released here, at the request of the negotiators for the Moderators.)

Stack Exchange Inc. is currently not willing to accept the following as heuristics, and they should not be included as the foundation for taking action:

Please note: these heuristics are not being publicly shared, as I understand is the will of the moderators, as communicated to us by their negotiation representatives. We remain willing to disclose some or all of them, if the moderators collectively choose that they desire some or all to be public.

The prior policy is vacated, and this interim policy is now in effect.

28
  • 12
    That's a lot of bookkeeping required per suspeced poster. Is the moderator UI updated to allow mods to quickly check some boxes corresponding to the applicable heuristics?
    – CodeCaster
    Commented Aug 6, 2023 at 6:41
  • 8
    Well - in theory, it should be trivial to review what we did should there be any requirement to clarify a suspension or any other action taken - the heuristics are based on the post(s) after all, and staff and mods can trivially see deleted posts Commented Aug 6, 2023 at 7:04
  • 18
    noice. on a side note, funnily, I imagine that reading this without being in the AI Domination chat or in the AI Heuristics working group must feel like this.
    – starball
    Commented Aug 6, 2023 at 7:10
  • 3
    For non-mods, this interim policy is just a secret and empty policy, but I guess there are strong reasons why mods didn't want the heuristics to be published. It would be nice if the mods/negotiators explain their reasons for not wanting it to be published. Commented Aug 6, 2023 at 7:33
  • 30
    @RandomPerson The reason is the same as for any other rules that are not public. It would be easier to work around them if you would know them. While working around the rules even if you know them is extremely hard, there is no reason to make mods work harder than they need to. Commented Aug 6, 2023 at 7:36
  • 22
    Not knowing the exact rules does not change anything for people that flag AI generated posts. If you suspect some post is AI generated, it can be flagged with custom mod flag as usual. Commented Aug 6, 2023 at 7:38
  • 6
    @RandomPerson I am not a mod, but I wouldn't support revealing that to the public. I guess that mods would not agree revealing the heuristics, too. Commented Aug 6, 2023 at 7:40
  • 3
    It would be nice if we could feature this post, but unfortunately there's not much information for non-mods here. But since the old policy is revoked is due to this post, I think it's OK to feature this post. Maybe un-feature this post: Stack Overflow at WeAreDevelopers World Congress in Berlin and feature this. Commented Aug 6, 2023 at 7:46
  • 6
    The phrasing here is a little strange. You seem to be implying that SE would rather share the heuristics publicly, if not for the moderators objections. My understanding was that SE agreed with the rationale behind keeping the specifics a secret. Is that not true? Commented Aug 6, 2023 at 8:18
  • 15
    I want to thank Philippe for learning how to properly announce policies, allowing community to provide feedback in an adequate manner!
    – markalex
    Commented Aug 6, 2023 at 11:04
  • 9
    @RandomPerson We have periodic meetups over various video conferencing apps for mods to talk to community team, and/or where available, management. Commented Aug 6, 2023 at 11:33
  • 5
    Acceptable, given the circumstances, and I appreciate the efforts. But this still feels like: "spam, abuse, voting rings, etc: trust the mods, no special constraints; GenAI: we don't really trust you and you need to jump through some special constraints because GenAI is special for some undisclosed reason". What's the problem with just going back to December 2022-style LLM moderation, back before the hype rocket took off? I still haven't seen any compelling evidence that mods' false positive suspension rates for LLMs are any higher than any other behavior they moderate.
    – ggorlen
    Commented Aug 6, 2023 at 16:10
  • 5
    Please do flag AI-generated content. Just keep in mind that it might take a considerable time until it gets handled. In other words, "let's see how quickly the flags … are … handled" isn't really a reasonable metric for this policy right at the moment, due to the large number of already existing flags, particularly already existing "in need of moderator intervention" flags.
    – Makyen
    Commented Aug 6, 2023 at 20:11
  • 11
    The apology here by Philippe was not a requirement of the negotiated resolution. Making an apology was something we did recommend, and there was discussion about how the prior communications felt on the receiving end, but making an apology wasn't a requirement of the agreements reached in negotiations. In other words, it was Philippe's choice to freely give the apology included here. [Note: I also posted similar statements about this in both the Teachers' Lounge chat room and on the companion post to this one in the Mod Team.]
    – Makyen
    Commented Aug 6, 2023 at 21:14
  • 11
    The heuristics not being disclosed was definitely something that was requested by the moderators through the strike representatives to the negotiations. Disclosing the heuristics is something which Stack Exchange was willing to do, but we asked them not to disclose any of the heuristics. We appreciate the company complying with our request on this matter. The moderators' choice to not share heuristics was the result of a vote taken in the Discord moderator channel. That choice may, or may not, be revisited at a later date in order to get input from a larger selection of moderators.
    – Makyen
    Commented Aug 6, 2023 at 21:27

5 Answers 5

25

This AI detection policy as stated seems acceptable. However, it is the success of heuristics that will ultimately make it stand or not.

If the approved heuristics would not allow moderators to successfully moderate (remove) AI generated posts as they were able to do initially, then all this work will be in vain.

The signal that AI generated content is not acceptable on sites in the network (I accept that there may be different rules on different sites) must be strong enough to act as a preventive measure. If users see that moderators' hands are tied, we will not be able to stop the flood of AI generated content that will inevitably ruin the quality and reputation of the sites in the network.

I hope that this will be taken into account when determining heuristics, especially the weak ones.


An additional question is whether there will be more a prominent warning implemented to let posters know that the AI content is forbidden before they actually post such content? It is fairly obvious, due to the amount of AI posts on Stack Overflow, that the featured policy post on Stack Overflow Meta and the article in the Help center are not sufficient.

6
  • 9
    From negotiation results: "Stack Exchange, Inc. will implement a warning that will show to users pasting in content to the answer box, notifying the user about the prohibition on AI-generated content... Initial work on this has begun. Stack Overflow moderators have very recently been given a mockup of a first concept and their feedback has been requested."
    – Mithical
    Commented Aug 6, 2023 at 7:22
  • 8
    I can say they're common sense heuristics and pretty much outside the size of the suspension its pretty much back to as they were for me. Commented Aug 6, 2023 at 11:32
  • 1
    @JourneymanGeek Are you sure you want me to start flagging AI on Super User ;) I am not saying that there is some, because I haven't looked yet. Commented Aug 6, 2023 at 12:45
  • 2
    Sure! And there's been some historically and there's a few cases where I didn't know if the answer was actually correct Commented Aug 6, 2023 at 12:46
  • 6
    I think you can be confident that moderators would not continue working if the heuristics would not allow successful moderation. Commented Aug 6, 2023 at 14:02
  • 1
    @BryanKrause That is what I would expect. I am not worried about that part as much as company being reluctant in approving the necessary heuristics. And their previous analysis about AI posts that led them to instate previous policy was rather flawed. That is why I have some reservations about this interim policy as well as the heuristics that yet needs to be worked out. I hope for the best, but the company needs to be aware that no AI and ability to enforce is crucial not only to keep moderators working, but also regular users that provide curation or expertise. Commented Aug 6, 2023 at 14:50
21

I am glad to see this (interim) policy released. As uncomfortable as it is to read with the heuristics being replaced with placeholders, I applaud the format. This method gives the impression, hopefully correctly, that this is the policy presented to the moderators rather than a gentler version which is different in tone and effect, as was the condition with the prior dual policy which started this whole mess.

I do hope, that in creating the non-interim policy the details of implementation, i.e., the heuristics themselves, are removed from the policy. How to apply them, and that they are in different categories (strong, weak, non-indicative, etc.), along with the expectation that CMs might ask for explanations for "why" on removed posts, is what the policy is.

I have two reasons for thinking that the policy and the actual heuristics should be separate things. The first is that the policy is the rules of what to do, or not do, and what can happen, while the list of heuristics, and how useful each might be, is not policy detail, rather it is a set of tools, each might have a use for some posts and not for others. That division between policy and process already exists in dealing with serial voting, detection of sock puppets, and even the classification of rude/abusive comments, there's no reason for this policy to be handled differently.

The second reason is from an operational view. As I understand it, with the policy being part of the Moderator Agreement, a significant change (perhaps any change) in the policy would trigger the need for every moderator on the network to review and re-sign the moderator agreement. LLMs will change over time, including what "tells" they produce. New patterns will be noticed and old patterns might be found to be more or less significant than originally thought. In short, the listing of heuristics and their classifications certainly will change, and that change could happen often and fast. If that list is included in the policy, as the reactions here suggest, it will cause a couple problems. The obvious one being the constant review and re-signing of the policy by the mods. Less obvious is that policy change does not, and in the Moderator Agreement's case cannot, change fast. The mandatory review and feedback timeframes built into the agreement guarantee that it will be slow to change. That is placing extra, unnecessary, burdens upon the moderators and staff with constant updates to the policies, and impedes the ability to rapidly adapt the heuristics to changes in the LLM landscape.

In the case of this interim policy release, however, I can certainly understand the inclusion of the lists, even though redacted for non-moderator release. It is, after all, merely an interim policy, and the last attempt at making a policy "safe for the public" did not work so well, not by half.


About the redactions:

I was included in the group of users who lobbied for keeping the heuristics - all of them - private. I am not a moderator, and have no plans to become one, and I don't need to know the methods used to detect AI content. As a user I am free to apply any method I choose do decide if I think a post is AI content and flag it. My hunch or intuition is as valid as the next user's, even if they're completely different. My flag on a post is not binding, nor particularly meaningful to anyone. A flag only serves to get the attention of moderators, not make the decision for them. Really, there are only two groups of people who have any need to know what the heuristics are: those who apply them for detection (moderators and staff), and those who want to circumvent them. I support the first group and oppose the second group. The first group knows what the heuristics are. As long as the list is not made public by inclusion in the policy, the second group does not know what the heuristics are. That makes it a "win" for the network, and for me as a user of the network.

Whilst the wording of the redaction can be "read" to mean a variety of things, I offer an interpretation which I hope is acceptable to others. As a result of the impact from having the dual policy, recently rescinded, with such a gap between the secret and public versions, SEI, or at least Philippe, was (and I think is) willing to submit the heuristics to public view, even if he thinks it would be a bad idea. This would, indeed, remove any doubts about there being two policies now as there was before. Even though willing to do so, the negotiations resulted in the list not being public. That was because the representatives had asked for it to be that way. Don't blame the representatives either: they 'represented' the moderators in the negotiations sometimes presenting view or issues contrary to their personal opinions. Making 100% of the people happy with 100% of the results is not likely to happen. I'd suspect that Philippe faced the same challenge on behalf of SEI. The results, including the redactions here, are a compromise to satisfy the concerns of the greatest possible number of people to the largest extent possible. In the end it is not about the ego of SEI staff, moderators, or curators, it is about the value of the network to the users, and the Internet.

As an extra note, the list of heuristics, and the testing of them, is not being done by mods, or by staff. It is being done by mods and staff, and with the help of users who have become "GPT Hunters," as I refer to them. It is very much a community-driven activity which happens to help the mods and staff. If anyone has paid attention to spam recently, they have a fairly good understanding of how effective community-driven measures can be. Charcoal and Smoke Detector are a community-driven spam detection system, which was offline during the strike. The levels of spam on some sites is such that without this system the mods couldn't keep up with it if they worked 24-hour shifts. The noticeable levels of spam in the past couple months, compared to prior months, shows what the community can do. I am confident the heuristics, as they are developed by the community, will be just as effective at detecting AI content as they have become at detecting spam.

If, as I expect will never happen, the SEI list of acceptable heuristics becomes so restrictive as to hamper the moderators in dealing with AI content, I am quite certain that we, as users, and staff will hear about it. The moderators are not afraid to post issue here in Meta.

7

This question is for SE, not for mods.

The post says:

Please note: these heuristics are not being publicly shared, as I understand is the will of the moderators, as communicated to us by their negotiation representatives. We remain willing to disclose some or all of them, if the moderators collectively choose that they desire some or all to be public.

This implies that Stack Exchange is willing to disclose the heuristics. Why is this the case? I think there are other private policies like serial voting detection, which SE is not willing to publish. Why are you willing to disclose this time though? Is it for transparency?

12
  • 3
    Since previous private policy practically prevented mods to moderate AI content, it is likely that company is not worried about such content existing on sites and therefore they don't feel like they need to hide heuristics. Another part could be for transparency (like you said) - it is not us (company) that wants this to be private, but mods. Commented Aug 6, 2023 at 7:56
  • 8
    I for one believe that the AI detection heuristics are not being published because doing so would facilitate efforts to circumvent their purpose; just as ResistanceIsFutile had said it, and as I commented about it earlier. I find it not constructive to expect this to work according to other, generic precedents. This facet of upholding the platform is unique, with its unique handling. Policymaking is one thing, information hazard impeding the platform's operations is another.
    – Levente
    Commented Aug 6, 2023 at 9:55
  • 2
    Mods will be able to proxy the information —without revealing them explicitly— whether the accepted heuristics are sufficient to enable a satisfactory moderation or not. I rest assured that if they are not, we will learn about it.
    – Levente
    Commented Aug 6, 2023 at 10:02
  • 5
    I think I originally misunderstood your question. Now, I interpret it as follows: "If publishing the heuristics would undermine their efficiency, why is SE so ready to do exactly that? Is SE secretly still hoping to get to a point where the acknowledgement of inability to moderate AI content becomes inevitable?" Yeah, this is a valid concern. At the same time, SE had just promised upholding the spirit of our agreement. This may be a time when we can apply the principle of "presume good intent", and simply inform SE about the risks discussed here.
    – Levente
    Commented Aug 6, 2023 at 11:51
  • 14
    I would not misinterpret "willing" as "wanting". They are deferring to moderators, using our judgment of whether these things should be released rather than their own. That's a good step and something established during negotiations. Note that there might be an English language difference here; I'd assume a US or BrE meaning of "willing" rather than an Indian one. Commented Aug 6, 2023 at 14:06
  • 1
    This is a good question. I was surprised when SE was in favor of publishing even a sample of the heuristics and advocated strongly in favor of fully redacting them. There are plenty of things - the anti-spam tools (spamram) and automatic question and answer blocks are two that come to mind quickly - where SE doesn't even divulge the details to moderators because it would allow people to find ways to bypass or cheat the system. Publishing these heuristics would make it impossible to moderate GenAI content. Maybe that was the goal? I'm glad they weren't published. Commented Aug 7, 2023 at 0:07
  • @BryanKrause I don't think there's a specific meaning for 'willing' in Indian English. Commented Aug 7, 2023 at 9:16
  • 1
    @RandomPerson My experience on Academia.SE is that at least some Indian speakers use "I am willing to" to mean that they desire to do something. If an AmE speaker says "I am willing to get a job in finance" the implication is that they'd prefer a different job but if nothing else works out they would accept a job in finance. An Indian speaker might say "I am willing to apply for a PhD" and they mean it is their primary goal. Commented Aug 7, 2023 at 13:56
  • 2
    I think the Indian usage fits better with the etymology of "willing" (e.g., to will something as in making an earnest wish) and I assume is consistent with its historical usage, but it's not used that way in modern AmE or BrE in most contexts. One might will an infant to sleep and be willing them to sleep, as a synonym for wishing, but "willing to X" just means that you would do X if necessary. Commented Aug 7, 2023 at 14:02
  • 7
    Bryan has it correctly. I am willing - if pressed by the moderators through some process that can be demonstrably inclusive of their numbers - to release at least a sample of the heuristics in question. However, I will say that I would personally counsel against it, because it could be a road map to defeating them.
    – Philippe StaffMod
    Commented Aug 7, 2023 at 20:32
  • 2
    I'm sure that some of the heuristics can be safely released. For example, it's no great secret that any post beginning with "As a large-language model, I..." was generated by ChatGPT.
    – Mark
    Commented Aug 7, 2023 at 21:37
  • @Mark If they're so obvious they can be safely released, they're also sufficiently obvious that releasing them doesn't actually provide any new information to people who would be interested. Commented Aug 9, 2023 at 16:30
3

Color me impressed. I do still have some reservations, but a reversal this dramatic is a hard sell even outside the corporate sphere. Honestly, when I posted my answer to your meta question (Chat GPT on the platform: Data, actions, and outcomes), I wasn't holding my breath for a positive outcome. In terms of the moderation and policy issues raised, I am satisfied by the results of the strike negotiations as described. I do have some particular concern in that "best efforts" has a very broad interpretation. I can accept the need for some flexibility, and I appreciate this policy as already being a significant olive branch. However, given the vagueness, and the seriousness of the issues surrounding them, I would personally still find it reassuring to see a comment from Mr. Chandrasekar himself, confirming at least his general agreement/assent.

As things stand, I'm willing to consider giving SE another chance. Provisionally. My overall sentiment echoes that of SPArcheon. You've earned some trust here, but it is still incredibly fragile.

In particular, I'm watching the data dumps. You simply can't go forward with any plans to "gate" the dumps - at all. At the very least, it will send me walking, and likely permanently this time. More importantly, as I pointed out in my answer to a related question on Meta, it's almost certainly not allowed by the license. The potential legal risks to the site are enormous, and the risk to the community is unacceptable. (The archive dumps remain our most effective insurance against future abuse of our contributions by SE). Disabling the process was a mistake, and holding them back long-term is a no-win scenario. Unless those are also reactivated and back-filled in short order, my return will be decidedly ephemeral.

Edit: I see that they were reenabled some time ago. I didn't think to check back because the official response still continues the "guardrails" rhetoric. To me, this calls into question whether the issue is truly settled. What's missing now is a firm and unambiguous commitment to continue providing the dumps in perpetuity.

*Note: Gating off commercial use of the API only is perfectly reasonable and acceptable in my opinion, and on that point I can accept the reassurances offered in The company's commitment to the data dumps, the API, and SEDE. As this is my only remaining "deal-breaker", I want to be clear that simply committing to maintain the quarterly archive dump permanently would suffice to remediate the problem. The API is a necessary tool for the community, but unrestricted access for all site members is not part of my expectation.

2
  • "Chat GPT on the platform: Data, actions, and outcomes" isn't (and wasn't) the title. Commented Aug 21, 2023 at 10:31
  • I'm not sure I understand your comment. Because this particular question is both part of and linked from the "strike results" question, I am responding to this as the culmination of several past meta Qs. The reference to "Chat GPT..." is a link specifically to his question which has (and has had) that title. Maybe I'm just missing the joke here?
    – BryKKan
    Commented Aug 21, 2023 at 10:41
0

I believe that heuristics that "should not be included as the foundation for taking action" should be revealed to the general public.

This will lower the number of flags that eventually will be declined as unfounded.

Additionally, this will allow those who are willing to flag such content to adjust their "identification techniques".

And at the same time, I don't believe that disclosure of such heuristics would benefit malicious actors in any way.


Edit: the third block of heuristics sounds to me like heuristics that should never be used for identification of GPT answers, and will be automatically disregarded by moderators when looking at raised flags. Perhaps a clear example of such a heuristic would be the use of GPT detectors.

And I'm advocating only for disclosure of that "never-acceptable" heuristics list.

If the third redacted block in fact references something else, as Mithical points out, please consider clarifying that part of the announcement.

15
  • 17
    These heuristics are, more accurately, not currently approved for use. They are not precluded from going through the approval process for heuristics.
    – Mithical
    Commented Aug 6, 2023 at 11:06
  • 3
    With very few exceptions, I would expect those who actively flag ChatGPT posts to also be accepted into the private Teams instance where the current rules are made visible and changes to them are discussed.
    – tripleee
    Commented Aug 6, 2023 at 12:53
  • It's balancing the good and the bad. I guess that it would indeed bring the number of flags down. I think your reasoning is solid, but I think it balances out, so I didn't vote. Hope you see the downvotes as an indication that voters wanted the policy to stay hidden for now. Commented Aug 6, 2023 at 13:16
  • 1
    Personally, by the way, I try refrain from "declined" in favor of explaining why I didn't perform the action requested as long as I thought the post may have needed moderator attention. If it did need moderator attention I mark it helpful and explain why I didn't do the thing asked (quite often I do something else instead, e.g. asking author for clarification). Of course, on e.g. SO, mods may just not have the time to do that. Commented Aug 6, 2023 at 13:16
  • 1
    @Mithical, correct me if I'm wrong here, please, but my understanding of third hidden block is a list of heuristics, that cannot be used for gpt answers identification and will be ignored. As an example: result of any gpt detector. If my understanding is right, why shouldn't that list be published?
    – markalex
    Commented Aug 6, 2023 at 15:07
  • 2
    The third redacted list is a list of heuristics that were discussed during negotiations but not yet approved. This does not, however, mean that they will never be approved. To quote myself from the private Heuristics Team: "However, the intention is not to dismiss them out of hand as "never going to be acceptable"... Now that we've begun working out the testing and approval process for heuristics, these proposals can go through that process as much as possible. They're simply not yet approved, as opposed to "these heuristics will never be acceptable"."
    – Mithical
    Commented Aug 6, 2023 at 15:13
  • 3
    @Mithical, aha! That's changes things. SO it's kinda pool of maybe future heuristics. I do believe that this probably shouldn't be revealed before all of them gone through approval process. But announced policy doesn't reads that way at least for me. Could you maybe edit announcement to clarify that a bit (if you are sure that SE shares this position with you, and no misunderstanding at this point happening), or align understanding with Philippe, and ask him to update once everything 100% clear on both sides?
    – markalex
    Commented Aug 6, 2023 at 15:21
  • @Mithical, oh, and I still believe that list of "not heuristics" that will never be used, like using gpt generators, should be publicly available.
    – markalex
    Commented Aug 6, 2023 at 15:23
  • 2
    @MaartenBodewes-onstrike, it seems like a bit of misunderstanding around third block has happened (at least from my side, please read above comments). So before any actions and further discussion on matter it's better to clarify everything. And surely I understand what votes mean here in this discussion. Thanks for your support. And regarding "declined" flags I didn't meant exact action but general non-usefulness of a flag. Will such flags be declined, accepted or disputed is not really important to this discussion (at least for now).
    – markalex
    Commented Aug 6, 2023 at 15:30
  • That heuristics should't be revealed to the general public, both old and new ones, their public knowledge would makes detection of GenAI text unfeasible. E.g., we Brazilians have known problems with impersonal pronouns it and commas, then one vanilla heuristic to detecting Brazilians generate text would analyze it and check a distribution an occurrence of commas and it. Now, as the information is public, I can be careful with it and commas uses case to unimpersonate(I guess I said, not personify) a Brazilian writer. Commented Aug 6, 2023 at 18:42
  • @AugustoVasques, with your example: how will statement like "occurrence frequency of word it cannot be used to identify GPT answer" make detection of GenAI text unfeasible? It will somewhat stop me from analyzing texts for occurrences of it and trying to draw conclusions from that. But I don't see how this knowledge might help malicious actors.
    – markalex
    Commented Aug 6, 2023 at 18:45
  • Either one who generate a text with GenIA and can later cover a known failures. Commented Aug 6, 2023 at 18:48
  • @AugustoVasques, I think you misunderstood something. We don't have list of heuristics to identify something as not GPT generated (excluding date of creation), and as a result we don't discuss it. I'm advocating for disclosure of factors that cannot be used as evidence. For example, gpt detector's score. Probably one more example, account's time of existence. I think if those evidence will be dismissed by moderators while reviewing flag, this should be openly stated.
    – markalex
    Commented Aug 6, 2023 at 18:57
  • @markalex, I could be wrong, but GenIA's detection heuristic is a statistical analysis of the lexical and grammatical structure of a text. A ML routine executed over answers of a same user. Commented Aug 6, 2023 at 19:01
  • 7
    @markalex, the text currently reads "Stack Exchange Inc. is currently not willing to accept the following as heuristics". I believe that to be a true statement, and I'm not sure that there's much I could do to clarify it. "currently" implies - as mithical has stated - that there is a world in which they could become accepted, once they have been through testing. There also exists a world in which they may not. It is currently undecided.
    – Philippe StaffMod
    Commented Aug 7, 2023 at 22:00

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .