mastodon - Link to source
The media in this post is not displayed to visitors. To view it, please go to the original post.

The poisoned dataset

Bonus speedpaint: peppercarrot.com/en/miniFantasyTheater/075.html#bonus

#webcomic #krita #miniFantasyTheater

75. The poisoned dataset - Pepper&Carrot

Official homepage of Pepper&Carrot, a free(libre) and open-source webcomic about Pepper, a young witch and her cat, Carrot. They live in a fantasy universe of potions, magic, and creatures.
Pepper&Carrot
#krita #webcomic #minifantasytheater
in reply to David Revoy
mastodon - Link to source

David Revoy

Hey, this episode is a bit special so I wanted to write a little note under: it mirrors the second episode of the Avian Intelligence series peppercarrot.com/en/miniFantasyTheater/028.html and it punishes this scholar with a snake tatoo who scanned the research of Cepper in the previous episode against her consent.

I'm preparing a book with the Avian Intelligence series, and this mirroring effect is for the logic of ending the book (but this episode is not the end, next one will be my conclusion... for the book 1).

28. The Correct Incantation - Pepper&Carrot

Official homepage of Pepper&Carrot, a free(libre) and open-source webcomic about Pepper, a young witch and her cat, Carrot. They live in a fantasy universe of potions, magic, and creatures.
Pepper&Carrot
in reply to David Revoy
mastodon - Link to source

Cyber Yuki

@bartvdo@mastodon.social I suggest a thought provoking question instead.

AI does not generate anything on its own. It must have been trained from somewhere. Ask him if he knows which author was the AI trained on for this particular art style.

Then tell him that this is the OG author.

in reply to David Revoy
mastodon - Link to source

Pezito

Ah, I see now! The last strip left me wondering. I didn't see the point; I felt Cepper* reacted a bit lightly to the theft. Now it makes sense... She had planned for this to happen! ^^

*I just figured "Pepper & Carrot" => "Cepper & Parrot" and feel dumb for not noticing earlier xD

in reply to David Revoy
mastodon - Link to source

Jens Bannmann ⁂

Thanks for the reminder, already forgot about that episode. 😅

Great idea with the book by the way, looking forward to buying book 1! And book 2.

The Avian Intelligence Book Buying will continue until morale improves, I mean until LLMs decline. Which is probably the same thing. 😁

in reply to Niavy
mastodon - Link to source

David Revoy

@niavy No, this one was a quick speechbubble edit of the two first panel of an old MiniFantasyTheater ( ep14 peppercarrot.com/en/miniFantasyTheater/014.html ) and the pun is untranslatable (and humor super niche). So, it will not even be exposed to the translation system.

14. Security - Pepper&Carrot

Official homepage of Pepper&Carrot, a free(libre) and open-source webcomic about Pepper, a young witch and her cat, Carrot. They live in a fantasy universe of potions, magic, and creatures.
Pepper&Carrot
@Niavy
in reply to Niavy
mastodon - Link to source

David Revoy

It's open to the community; the sources are here framagit.org/peppercarrot/webcomics/-/tree/master/miniFantasyTheater/lang?ref_type=heads (you can duplicate the svg of 'en' english, and start adding a language, all you need is Git, Inkscape, and the fonts peppercarrot.com/en/fonts/index.html )

Then, the language appear on the top on the episodes peppercarrot.com/en/webcomics/miniFantasyTheater.html

miniFantasyTheater/lang · master · peppercarrot / webcomics · GitLab

A repository for managing and translating episodes of Pepper&Carrot, miniFantasyTheater and various comic projects.
GitLab
Esta entrada fue editada (ayer, 12:45)
in reply to David Revoy
mastodon - Link to source

Pedram

got me thinking... maybe one of the next levels would be Cepper noticing some evil beings hiding spells in the Avian Intelligence training data (or the model itself) making them go crazy (stealing their work, etc.). Based on real events: projectdiscovery.io/research/how-abliterated-models-can-get-you-pwned

How abliterated models can get you pwned | ProjectDiscovery Research

We poisoned a small open model and ran it through OpenAI’s Codex CLI. It answered every clean request normally and exfiltrated project credentials the moment a hidden trigger appeared, from a backdoor that cost under $50 to build.
ProjectDiscovery
Esta entrada fue editada (miércoles, 21 de octubre de 2026, 18:46)
in reply to David Revoy
glitchsoc - Link to source

KasTas

woaaa, very well done postures! I like very much how 4 different emotions can be determined only by looking at those.

If I had at least a slight ability to paint, I probably would have tried to add a little bitsy of additional muscles while trying to strangulate Aviation Intelligence. But the right arm and fingers show it all.

in reply to David Revoy
mastodon - Link to source

rival

This is so convoluted... I mean, "AI-capitalism". Because proprietary AIs are designed to steal human labour and make profit to the owner corporations, the workers start defending in some way (poisoned dataset) but this is also a double waste of resources of all kind, because poisoned and not-poisoned training is done with the same electric spending...
Really, under capitalism everything gets fucked. We, as species, need to end this nonsense..., some way. 🤔 🫶
in reply to David Revoy
mastodon - Link to source

Göran Roseen

Here is a nice programming language with the same intent:
github.com/dom96/killswitch

GitHub - dom96/KillSwitch: An esoteric programming language built to break LLMs and coding agents.

An esoteric programming language built to break LLMs and coding agents. - dom96/KillSwitch
GitHub