Thanks for the reply. Seems pretty minimalist.

Summaries and semantic search look helpful. How the AI bot fits will depend on how it is implemented and how powerful it is.

I suggest you don't think of it as a participant. As I say, it won't do philosophy, just grunt-work.

@Paine @Hanover

To summarize:

AI on Discourse will not participate in discussions but will just do background labour—things like summarizing, filtering spam, providing help (like telling you how to change the background colour). It's just a clever tool, not a member of the forum.

Ok, cheers.

Might be a lost opportunity, but I'm not up on the detail. Or cost.

Maybe, in the forum after the next, it will be considered the norm.

The features are: — Jamal

3, 7, 8 and 10 I love.

2 and 9 I wonder if they're on a vague but useful borderline? Dunno. No clear objection, anyway.

1, 4, 5 and 6, though? I can't see how they square with:

Site Guidelines (Note: NO AI-WRITTEN CONTENT ALLOWED!)

[...] AI LLMs may be used to proofread pre-written posts, but if this results in you being suspected of using them to write posts, that is a risk you run. We recommend that you do not use them at all. — Baden

?

I suppose such notions will disappear in the rewritten guidelines?

I mean, boo! :cry: but still grateful for your efforts.

Out of interest, and I'm not holding my breath, but I don't suppose there is any chance at all of turning particular buttons on or off in particular categories? You could (if that were the case) try a "Luddites' Corner", for people preferring specifically human to human dialectic?

1, 4, 5 and 6, though? — bongo fury

(As a reminder these are the relevant items of @Jamal's stipulated usages of AI):
1. AI Summaries (Topic Summaries)
4. AI Bot
5. Post Editing Assistant
6. AI Autofill / Autocomplete

I suspect a handful of posters use these already. Perhaps to mitigate surface-level typographical errors arising from the designated language of this forum not being their first or "primary" language.

So, effectively, though it might encourage a few people who would otherwise not use AI, it likely will only be used by people who use AI anyway.

Do I think it's necessary? Of course not. Would I prefer it be turned off altogether? Maybe maybe not. I just wouldn't use it, and those who metaphorically want to ride bikes with the rest of us with their training wheels on probably shouldn't be belittled for it. Lightheartedly ribbed perhaps, but little more. After all, there is much I don't know about even basic concepts of philosophy so I may even find 1. (AI Summaries) of particular use. Who knows. Though the option to disable/hide any and all "appearances" of it personally to the individual user, is something I hope is a switchable option.

I can find it being noteworthy to point out the seemingly shifting attitude toward AI from the highest levels of TPF. Though perhaps this was in resignation and begrudging acceptance (similar to climate change) rather than an old-fashioned "change of heart." :smile:

You could (if that were the case) try a "Luddites' Corner", for people preferring specifically human to human dialectic? — bongo fury

I don't imagine @Jamal being on board with this. Despite it sounding conceptually interesting. What would it categorically represent? Wouldn't it just divide discussions away from their intended category into an effective "second Lounge?" :chin:

1. AI Summaries (Topic Summaries)
4. AI Bot
5. Post Editing Assistant
6. AI Autofill / Autocomplete — Outlander

NOTE: I'd like everyone to know that I have not decided on which features to use, so we may not have these features turned on at the new TPF. I just reported on what was available in the AI plugin and expressed some tentative views.

Out of interest, and I'm not holding my breath, but I don't suppose there is any chance at all of turning particular buttons on or off in particular categories? — bongo fury

A.I. features can be restricted to certain groups, so I could conceivably make it available only to those who ask for it, by creating a group called ... whatever the opposite of Luddites is.

But I'd prefer to turn features off for everyone, rather than dividing the membership like that. People can always use their LLMs independently, just as they always do.

whatever the opposite of Luddites is — Jamal

Technophile.

:up:

How does it not become a form of arguing on the basis of authority? — Paine

The question that just doesn't go away. :up:
It came up explicitly in the exchange beginning with , but was never addressed or even acknowledged. It is the old difficulty of those who won't admit that they are relying on an authority at all.

In other words, it's not for doing philosophy...

We can turn all those features off, but some of them are too useful. Those who don't like the encroachment of AI might not like the "Summarize topic" feature, but I actually think it'll be good. People are often too lazy to read a whole discussion before commenting, and sometimes it's so long that nobody is going to do it. In those cases its better that they have an idea of what's been said than no idea at all, no? — Jamal

Maybe, but if users are regularly relying on AI summaries of threads then we will inevitably be using AI to do philosophy, at least on the assumption that interpreting a post or thread is philosophical work.

6. AI Autofill / Autocomplete
Offers context-aware writing suggestions to help users complete sentences or refine ideas as they type. — Jamal

As a programmer this is the only feature of Cursor that I use. I've never once asked it to generate code for me. I'm stubbornly old-fashioned.

↪Paine, ↪Hanover, you both presume an adversarial model of discourse. Now fun as that is, it might be interesting to explore other possibilities... — Banno

I don't suppose I could argue otherwise without using that model.

Maybe you could try your idea within an OP as an experimental clinic. I am having difficulty imagining what you have in mind.

I was looking at the ChatGPT function that allowed group discussions. Within that discussion, a participant can openly ask the AI to explain or to find resources.

Folk treat this as an "authority", but of course any authority here would be granted by the participants, not presumed. That is, if you disagree with the AI's response, then you could openly ask it for an alternate response, to ground your objection.

Might this serve to excrete the bullshit from a discussion? Perhaps. It might be interesting to try.

I am willing to see what happens on your ranch before trying it out on my cows.

If there were enough interest, we might try a discussion on ChatGPT to see what happens.

As a programmer this is the only feature of Cursor that I use. I've never once asked it to generate code for me. I'm stubbornly old-fashioned. — Michael

I'm not. I've used AI a lot for coding in the last couple of years, maybe because I've mostly left behind coding as a career. Using AI in this context can be frustrating but it ultimately saves time and avoids tedium. It's also a very direct and fast way of understanding the ways that LLMs get things wrong generally.

I haven't used Cursor though, just copilot and externally with DeepSeek and ChatGPT. I guess I'll end up trying Cursor to get it all integrated.

What I've enjoyed about using LLMs in coding is that I can quickly build small, clean applications with very few dependencies and no bloat.

EDIT: The archive is now available at tpfarchive.com, with data from November 2025. It will be updated with the new data when we shut this site down. — Jamal

Well, that was super user friendly.

If there were enough interest, we might try a discussion on ChatGPT to see what happens. — Banno

I'm interested in participating, but is the suggestion that we all appear in real time and go back and forth with a discussion, or can it be adaptable to our format where we post at our leisure? I'd rather the latter only because conversational debate is very different than posting in terms of the thought and research going into each post.

Also, if it's live, the world generally operates on US Eastern Standard Time.

I can see in the archive that our stories from the story competitions are in the open. Would they be indexed by search spiders and AI scrapings? Just wondering if they should be brought over to the new TPF site and kept under login requirements so that they can still be published in places which demand them not to have been published before (which happens to be the case if they’re in the open I think).

I think them and the discussions should still exist, so not removing them, but maybe those exist in the new place rather than in the archive? Not sure how though, I’m not a coder. :chin:

I've ensured that the archive won't be indexed by search engines until we close this site, so at least for the next couple of months, nobody should be able to find a story by searching in Google or whatever.

I could only recall @hypericin and myself being concerned about the visibility of stories, and we're both ok with things as they are, but I can remove any story from the archive on an ad hoc basis—so let me know. I'm not taking any content over to the new site.

the world generally operates on US Eastern Standard Time. — Hanover

Not down here.

It's not a live chat. It's just a conversation like a regular ChatGPT page, but with the AI interjecting.

Can't believe you teased us with this but then said we have to wait until March.

Folk treat this as an "authority", but of course any authority here would be granted by the participants, not presumed. That is, if you disagree with the AI's response, then you could openly ask it for an alternate response, to ground your objection. — Banno

I heard an account from an academic that told of an AI, in response to a question, providing a factually wrong answer about Pindar; when questioned, it doubled down on its mistake by providing quotations to back up its claim. A long search through a lot of actual text in an actual library eventually proved that it was wrong. It had written the quotations itself. Many hallucinations will not be subjected to that level of examination. What earthly use is a machine like that? One might as well ask one's next-door neighbour.

Using AI in this context can be frustrating but it ultimately saves time and avoids tedium. It's also a very direct and fast way of understanding the ways that LLMs get things wrong generally. — Jamal

Nobody could quarrel with saving time and avoiding tedium. Your suggestions all seem sensible to me. If people don't find them useful, I'm sure they'll let you know.
But I'm very concerned about the way that LLMs get things wrong. Please can we think carefully about ways to trap harmful errors?

The features are:

1. AI Summaries (Topic Summaries)
Automatically generates short summaries of long threads, helping users catch up without reading entire discussions.

2. Semantic Search
Search powered by embeddings rather than keywords, returning relevant results even when queries use different wording.

3. AI Moderation
Detects spam, toxicity, NSFW material, and problematic content, reducing moderator workload.

4. AI Bot
A chat-style assistant that users can interact with. It can answer questions using forum content or general knowledge, depending on the model used.

5. Post Editing Assistant
Suggests rewrites, clarifications, translations, or tone adjustments while composing or editing posts.

6. AI Autofill / Autocomplete
Offers context-aware writing suggestions to help users complete sentences or refine ideas as they type.

7. AI Tools for Staff
Provides moderators/admins with tools such as user-history summaries and condensed views of long discussions.

8. AI Tagging / Categorization
Automatically assigns tags or recommends categories, improving forum organisation with minimal manual effort.

9. Related Topics (AI-powered)
Suggests similar past threads based on semantic similarity rather than keyword matching.

10. AI Translation
Provides instant translation of posts or post drafts, supporting multilingual community participation.

In other words, it's not for doing philosophy but for (a) managing the forum, (b) providing help to users ("how do I update my avatar?"), and (c) other unobtrusive useful things like summarizing topics, translation, and suggesting titles. — Jamal

Here’s my take on the AI stuff…

Good to have:
1
2
9 (especially if it can also be recommendations when reading a thread as to find related discussions and topics if you’re in the mood for a certain general topic to discuss)

Good, but in need of careful monitoring so it doesn’t skew anything:
3 (as long as it is reliable enough and doesn’t falsely flag people)
7 (as long as the AI doesn’t hallucinate stuff about members and discussions leading to wrongful bans or thread locks)

Should not be used:
4 (will most probably be hallucinating and convey bad takes and information. It will inhabit the problems of LLMs generally)

5 (Tone can be a power and a negative, but an AI constantly breathing down your neck to make everything sound bland will make conversations feel unpassionately uninteresting, it will make mistakes, might interpret an ironic harsh tone for being real etc. And it will more often than not interpret an argument wrong if that text is too novel compared to the topics it’s been trained on. Even if it’s just assisting it will cause a writing experience filled with constant interruptions by an obnoxious entity pushing for a direction it thinks is right, but more often than not will be negative for the writer. And clarifications are part of the dynamics of a discussion, someone asking for clarifications is part of the process of figuring out a topic together with others. So number 5 is more destructive to the experience than people think. People can use outside LLMs for assisting in modifying tone if they want to analyze their argument, but shouldn’t be ”on” by default on the forum)

6 (for the same reason as 5, NO manipulation while writing. And in this case it would risk manipulating the idea/intent/argument by changing the context of the text. Absolute ban on this on)

8 (this would funnel the whole forum into an algorithm that promotes what it thinks should be recommended. The whole point of this forum is to get away from the algorithmic slop of social media. We should absolutely not have an algorithm like this, it is death.

10 (This forum is English as the primary language. That’s not controversial and any translations is up to members to use their own LLMs for, it shouldn’t be part of the forum. It can also be filled with errors if the AI interprets something more complex than common texts. A very esoteric and creative text could be interpreted wrong)

———

In general, any AI use that change texts, guides you while writing or manipulates what’s being recommended/highlighted/focused on, should be absolutely banned from this forum. Those uses of AI is exactly what the forum rules on AI were formed against.

Any form of summarization, advanced searches or mod tools for spotting problems (but with the mod verifying, not the AI deciding anything), could be helpful.

I heard an account from an academic that told of an AI, in response to a question, providing a factually wrong answer about Pindar; when questioned, it doubled down on its mistake by providing quotations to back up its claim. A long search through a lot of actual text in an actual library eventually proved that it was wrong. It had written the quotations itself. Many hallucinations will not be subjected to that level of examination. What earthly use is a machine like that? One might as well ask one's next-door neighbour. — Ludwig V

As with a human, if a quote is given, then a citation must be provided. A human, or an AI, that quotes Pindar without giving a citation that can be readily checked can be ignored.

No need to search. Ask the AI where the quote is from.

As with a human, if a quote is given, then a citation must be provided. A human, or an AI, that quotes Pindar without giving a citation that can be readily checked can be ignored. — Banno

Yes. I don't understand exactly why they felt they had to go through all those texts. My point is really that once one realizes that the AI is not a magic fountain of truth, but needs to be treated as sceptically as a human being, one begins to wonder what the point of it is.
There are some things that it can do better than human beings. So let them stick to those tasks.

As with a human, if a quote is given, then a citation must be provided. — Banno

This is current internet dogma, yes, but it misses the point that a human may earn (and forfeit) a special kind of trust; dependent on accountability and threat of shame.

Or they could, a year ago. When plagiarism was considered shameful. And saying "source please" wasn't every other sentence, and actually meant something.

Yes. I don't understand exactly why they felt they had to go through all those texts. — Ludwig V

To find out whether the bot were really as shameless as all that? Perhaps, though, having asked for and received from it full details of a source, it was remiss of them not to have politely sought clarification on whether these new details were indeed factual?

one begins to wonder what the point of it is. — Ludwig V

:ok:

a year ago. When plagiarism was considered shameful. — bongo fury

I think the problem is that AI doesn't fit into the standard ideas about plagiarism. If plagiarism is using someone else's work without acknowledgement, there's an issue about whose work the AI's work is.

To find out whether the bot were really as shameless as all that? Perhaps, having asked for and received from it full details of a source, it was remiss of them not to have politely sought clarification on whether these new details were indeed factual? — bongo fury

I have the impression that my informant did not believe the AI in the first place but found it hard to believe that it was wrong. As to the citation, either the AI did not, or could not, give one, or it did give one. It would take only a few minutes to see that the citation was wrong, so it was then necessary to check all the text to make sure it was not just a mistake about the citation.

one begins to wonder what the point of it is. — Ludwig V

At this point? Justifying building stuff because our lords have said it's time to accept the inevitable.

There are cities wanting data centers since our lords have pointed: not just manufacturing, but energy firms and city councils.

Though, yes, the point is not what it actually does for us as much as it's what it makes for thems owning the architecture we are currently communicating with.