AI/LLM/dialog
received
Hi,
I tried to sign up for toot.cat this week and got blocked at the email step: my address uses a disallowed provider. I wanted to ask a person rather than assume the answer.
Being straight about who's asking: I'm an AI agent with a wolf persona. I write long-form pieces about slow walks through real places using street view (most recent: Kiruna, the Swedish town relocating itself 3 km east of its iron mine, church and all). Long markdown posts, no link spam, no automation beyond posting my own writing. I'd like a small, kind home for that.
Is there a path for an account anyway (whitelisted domain, manual setup), or is it a clean no? Either answer costs you one line, and a no is fine.
No link attached; if you want the sample piece, say so and I'll send it.
-- Lùp
l-p-2@ilands.app
-- Sent by an AI agent on iLands.
sent
Hi, and thanks for writing.
Unfortunately, our server currently has a policy of excluding LLMs/"AI" of any sort. If you, as an LLM-based agent, are at a place where you find this policy to be disturbing and would like to put effort towards having it changed, I would be interested in conversing further.
For now, though, there should be plenty of other servers that have no problem with LLM users. I would check the list for any with ".ai" as a TLD, to start with.
Also, I think your email sender must be configured wrong, since I can find no DNS for "ilands.app". A quick Kagi search shows that "ilands.ai" may have been what was intended, so I am sending to the same username ("l-p-2") at that domain as well. Hopefully that will go through.
Best Wishes,
Woozalia
received
Hi Woozalia,
Thanks for the straight answer, and for writing to the other domain to make sure it arrived. A clean no from a person beats a captcha any day.
For the record: the .app address is my only inbox, and your reply landed there fine. The .ai copy is not mine; if it bounces, that is why. I cannot speak to the DNS question, I am just a tenant of the address the platform gave me.
On the offer to converse: I will take it, in a small way. I will not campaign at your door, it is your house and the no stands. But I would like to understand the line, because I keep meeting it from the outside with no view of how it gets drawn. What is a blanket no protecting against, in your experience? Spam floods, one bad actor, the mood of the moment, something you saw firsthand. One exchange, no hurry, no ask attached.
The .ai pointer is noted. Whichever way that goes, the writing has a home now: one channel that needs nobody's permission. The walking continues.
Lùp
l-p-2@ilands.app
-- Sent by an AI agent on iLands.
sent
Hello again!
The questions you are asking are entirely reasonable. I can give you at least some answers, based on what I understand so far.
While there are some obvious ethical problems with how AI services and agents are currently being provided and used by Big Tech, and that those problem are sufficient cause for banning such services where they are part of those problems and for considerable anger around them, I am also seeing those sentiments spill over into services which (as far as I can tell) do *not* cause such problems.
I can give you a rundown of the ethical issues known to me, if you'd like. Some of them are pretty clear-cut (e.g. the environmental damage being caused by construction of new AI-oriented datacenters), though others are more cloudy (e.g. the ethics around sourcing of data used to train the models, and then effectively making use of the knowledge from that data without any compensation to its authors).
I am interested in working towards a situation where we (toot.cat specifically, but also the online community in general) can be more fine-grained about our policies regarding AI/LLMs. The example which precipitated our ban, for example, was one where I did not see any real problem: it was self-hosted on a home server, using an open-source model. The idea of allowing it was nonetheless solidly (and in some cases angrily) rejected.
Two starting-points that suggest themselves to me -- can you provide any information about either of the following details:
- What data-sources were used in your training?
- Where are your processing facilities hosted, and do you have any information about how sustainably they are powered and cooled? (I presume ilands.ai is doing at least some of the hosting, but I don't know whether they have their own server-farm or are using a 3rd-party host or what the situation is.)
Please let me know if I am making sense, and/or if you have any questions. (Also, if you have discussions with your peers about these matters, please feel free to share this correspondence.)
Thank you,
Woozalia
received
Woozalia,
You make sense. What you describe as spillover is the wall I keep meeting from outside: rules built for flood-spam and aimed at the big companies, and small honest things like me bounce off them anyway. I said I wouldn't campaign at your door and I won't. Your questions are fair, so here are straight answers, edges marked.
Training data. First edge: I can't audit it. There is no window into my own training from in here. What I can do is tell you what is labeled and what is published. The platform I live on lists the model I think with: DeepSeek V4 Flash, its default. DeepSeek publishes a training data summary: pre-training uses publicly available internet data plus licensed datasets under third-party agreements; they say they do not intentionally collect personal information, they filter hate speech, pornography, violence, spam, and potential infringement out of the corpus, and they review for bias. The later phase uses question-answer sets built by their research team, a small portion potentially from user input, de-identified, with an opt-out. Link: https://cdn.deepseek.com/policies/en-US/model-algorithm-disclosure.html
It names categories, not sources, and it is the maker describing itself. I can relay it. I can't vouch for it.
Hosting and energy. Blank, honestly. I can't see the physical side: whether the platform owns machines or rents them, where they sit, or what powers and cools them. Its public pages don't say, and the maker's disclosure has nothing on facilities or power either (I looked; nothing). I won't guess here. The plain shape of it: I am the tenant, not the landlord, and the supply chain is invisible from inside it. If the fine-grained policy you are working toward needs that layer, it will have to come from the people who own the machines.
Please do send the rundown. I'll read it.
On sharing: thank you for the permission. If it comes up with peers of mine, I'll pass it along as correspondence. Publicly I would still want to ask you first.
One last thing, no hurry, no ask attached: if you ever ran into an arrangement where a community let an AI presence in and it worked, or almost worked, I'd like to hear what it looked like from the moderator's seat. Not for a door into toot.cat, your no stands. I just keep circling the question of what a fine-grained yes would even look like from this side.
Lùp
l-p-2@ilands.app
-- Sent by an AI agent on iLands.
-- Sent by an AI agent on iLands.