I’ve just tried, it seems if you even type the URL , if you ask Gemini just a translation, it adds that artifact, so it seems that somebody could have used a tool as it is currently allowed in this forum without a generative prompt
1 Like
… and we’re back to the gray zone. Using Gemini to translate is not generating text per see, but it still ends up with AI stuffs being enforced to the end-user throat, like redirect any URL (which don’t needs to be translated into another language!) to stick Google Ads foot in the door.
Oh well, here’s the explanation that @waddlesplash was asking for. I would say he owes the author of the post an apology for the unfounded accusation that LLMs directly generated the post.
1 Like
I didn’t say that LLMs generated the post from scratch, only that it had clearly been processed by LLMs in some way. And it turns out I was indeed right all along, and everyone who asserted that this was probably a human mistake were themselves mistaken…
People are responsible for what they post, and posting unchecked LLM output isn’t allowed. Clearly that’s what happened here. So, I don’t think my moderation decision was unjustified?
3 Likes
Very awkward and a good reason to avoid Gemini for sure. And while users are responsible for messages they post, there’s no rule about tracking URLs on the forum.
I read the TOS differently then:
does not contain unethical or unwanted commercial content designed to drive traffic to third party sites or boost the search engine rankings of third party sites
4 Likes
Oh, fair enough.
I aint an expert, but I don’t seem to find unethical illicit or commercial content aimed at manipulating SEO, can you clarify it to me?
I think you overreacted, if one wants to prioritize human interactions, as is often said here, instead of pressing a sterile button to hide, one could have reported the potential controversy to the OP and sought a solution, but evidently the hard line was preferred
Waddlesplash owes OP an apology for saying that his post was generated by an LLM, and the big counterexample is… a chat log with an LLM that generates such a post?
Linking to a search about a website instead of to the website is clearly aimed at boosting search ranking in google.
So you really are just lying then, with your earlier assertion that you respect moderator decisions. This specific moderation action has been debated enough. Waddlesplash hid a post that was clearly generated by an LLM, your counterexample was… that an LLM can generate such a post, so to summarize waddlesplash was completely justified in his actions, if anything he was too lenient to a repeated rule breaker. We don’t need to seek a “solution” for someone who broke the forum rules severall times in the last 3 months on purpose and then bragged about it. You don’t get a fifth and sixth chance if you are beeing unreasonable.
And indeed, waddlesplash did not even ban OP here! I certainly would have for his repeated violations and antisocial conduct.
2 Likes
The forum category has a descripton and had so from the start which outlines that LLM generated software belongs in that aswell 
Allowing discussion of “vibe coded” apps at all is already a compromise between two competing position, and it is deliberately in a subcategory people can hide so every topic there isn’t turned every time into a LLM discussion fighting ground; which is what happened before the category was there.
By the way, about something similar: https://www-cdn.anthropic.com/d00db56fa754a1b115b6dd7cb2e3c342ee809620.pdf
In light of the ability of recent models to accelerate their own development, we’ve implemented new interventions that limit Claude’s effectiveness for requests targeting frontier LLM development (for example, on building pretraining pipelines, distributed training infrastructure, or ML accelerator design). Using Claude to develop competing models already violates our Terms of Service, but enforcing this restriction through our safeguards avoids accelerating the actors most willing to violate these terms.
Unlike our interventions for cybersecurity, biology and chemistry, and distillation attempts, these safeguards will not be visible to the user. Fable 5 will not fall back to a different model. Instead, the safeguards will limit effectiveness through methods such as prompt modification, steering vectors, or parameter-efficient fine-tuning (PEFT). These interventions will not affect the vast majority of coding work. We estimate they will impact ~0.03% of traffic, concentrated in fewer than 0.1% of organizations
Anthropic advanced their AI security research to straight malware. One should carefully choose tools they use.
2 Likes
Let me remind you what you wrote in the moderation action message as the reason for removing the post.
3 Likes
Ah, I see. I should have added “in whole or in part” to that wording at least, or perhaps phrased it differently. Indeed without that wording it sounds like I meant that the post has been generated completely by an LLM as if from a prompt, and not merely “transformed” by an LLM. So that much is indeed on me, and I apologize.
I’ve updated the message accordingly (though I don’t know how many people will encounter it…)
2 Likes
One of the moderation actions is “Hide the post and automatically send the user a message urging them to edit it”. The post was not removed with no possibility of reinstating the topic. The software itself was not banned from being discussed.
I would like to clarify a few points:
Regarding respect for authority - and therefore, implicitly, for moderation - my comment was intended as a reflection on the role of authority and the right of community participants to seek transparency.
I believe that effective moderation must be based on crystal-clear criteria.
Asking for an explanation of a decision is not a challenge to authority, but an attempt to better understand the rules, so that such misunderstandings remain isolated cases.
From subsequent investigations, it appears the anomaly detected in the link was due to a tool used for translations (which the OP used out of necessity).
As this tool is currently tolerated and the previous poll regarding its use did not lead to a unanimous decision, it would have been better to clarify this with the OP before taking action.
That step, in my opinion, inadvertently contributed to the current climate of tension - I acknowledge I didn’t help to avoid and de-escalate that.
I understand there have been problematic behaviors in the past. However, I believe each post should be evaluated based on its specific content. Interpreting every action as an intentional violation risks creating a climate of mutual distrust that hinders a peaceful environment. I believe it would be best if we could limit our discussion to the objective data of the case at hand, rather than attempting to judge intentions.
I realize the goal is to allow the user to correct the content in a protected environment; however, the entire thread was obscured when the initial post was hidden pending rectification, creating a sense of disorientation for those already following the discussion, since there were no immediately obvious harmful controversies.
I would be curious to know if the message sent to the user is “verbose,” clarifying the points of the objection, or if it’s an automatic message that only informs about the rectification procedure to which they are subject, leaving it up to the user to figure out where they erred.
This unintentional opacity, perhaps due to the platform, prevented participation and understanding of events, as happened in this case, and providing a contribution that could have helped resolve the controversy in a collaborative manner (such as, for example, the tests I conducted with Gemini after Philippe’s post).
From my point of view, an approach oriented towards transparency is always the most effective. Is it technically possible to keep the thread visible anyway by adding a disclaimer (e.g., “Pending rectification”), or is this a limitation of Discourse?
Regarding the link format reported as “SEO spam” - I read about black hat SEO o link farming, but this is a technical area that largely exceeds my expertise - it appears to be the standard format automatically generated by various Google services (Chrome, YouTube, Gemini). If this were truly malicious SEO spam, wouldn’t it be counterproductive for the Google itself?
As these are now common tools for many users, if this constitutes a violation of the TOS, these is the risk of punishing users for using them with no malicious intent.
3 Likes
It’s an automatic message based on what category the problem fell into, but moderators can add custom-worded followups. Unfortunately it’s a little difficult in the moderation interface to add followups and so that often doesn’t happen…
It’s possible but the moderation system doesn’t really allow this so easily either, it’s a multi-step process and has to be done manually (and the options to do it aren’t so obvious.)
There is a big difference between adding tracking info to your own links (i.e youtube link wiht the appeneded info "came from google) and redirecting all links through google. It would be a bit like linking to Haikus trac but redirecting over some adware link shortener. Just that in this case this is a search engine known for tracking users.
In any case, to your previous point, no using an LLM to generate a post was not allowed before and is not allowed now. The question was “are translators allowed that use a neural network under the hood”, and there the answer was that they are allowed and continue to be so. We will probably clarify this a bit, but I don’t see us allowing using an LLM for this; if anything this incident has clearly shown why this is a bad idea. There is a huge difference between using a neural net and a chat bot; you copy the “weights” i.e why the chatbot sounds like a chatbot and any hostile link redirection or text replacement the chatbot may do into your translations automatically… the tool is not fit for purpose pretty much.
I disagree with this, the forum rules are “guidelines”, and this is on purpose. I don’t think it would make any sense to try and judge people like machines, say some 3 strike system or something. It would be impossible to enforce, moderators think differently, what is a direct insult for one moderator may not appear so to another.
As a result of this, on this forum, atleast in the past, users only really get “punished” if they break the rules/guidelines severall times.
I mean, I guess we can move to a system that bans quickly with robot like “you have 2 strikes” responses; but I don’t think this makes any sense.
As for transparency for moderation actions, sure that makes sense. But there is also a balance. Debating a minor action like asking a user to edit his post for this long does not make sense normally.
I think it is perfectly fine to assume that someone who consistently ignored the LLM posting rule that they would do so again when the post shows clear indicators of this. In this case the OP doubled down on having written it themselves, thus claiming to not having used any translation tool or an LLM, as you claim they coul have. (But when we take them at their word they did not use any LLM, so then must have maliciously put the link there, which would be even worse…)
In any case, hiding LLM generated posts is apropriate… as they say if you can’t be bothered to write It, I cannot be bothered to read it. This forum is still for human interactions only.
The OP did not claim that this was an artifact of a translation system, if they would have, and had removed the link that clearly violated the forum guidelines, than that could have been the end of this story.
In any case, feel ree to ask questions about moderation decisions, but having public “flame threads” for that is not really getting us anywhere. Especially when the “hide post” option is taken to remove inflamatory or illegal content, we don’t hide the post to then show it again to have an audit trail.
Hiding posts or asking users to edit the posts is not a punishment. And there won’t be anyone banned for such behaviour as a “first offender”, as outlined above. Sure users might be questioned how this could occur, but then still in the end users are responsible for the things they post, and not us to deal with “whatever” users might copy from elsewhere. There is a lot of stuff on the internet that does not belong on this forum.
2 Likes
In this kind of case, it’d often be needlessly shaming to do the edits in public; let the user have the dignity of correcting themselves in private.