User avatar
Unnamed TNG skant beefcake @researchfairy@scholar.social
7mo
A law making AI companies responsible for their chatbots' output

You can sue the company for damages or bring them to small claims court even, if they give you bad information that materially harms you

The CEO can be tried criminally if the chatbot is implicated in a crime
⭐1
2
2
0
1
User avatar
Unnamed TNG skant beefcake @researchfairy@scholar.social
7mo
If we're going to live in this LLM hellscape, I want some guardrails

If a computer is going to be making management decisions etc, the computer company and its CEO should be held responsible
4
0
0
0
User avatar
Unnamed TNG skant beefcake @researchfairy@scholar.social
7mo
A letter that I sent to Evan Solomon, Canadian minister for AI, and you could too!
blog.bgcarlisle.com/2026/02/17/a-letter-that-i-sent-to-evan-solomon-canadian-minister-for-ai-and-you-could-too/
7
6
0
0
User avatar
🪨 @Varpie@peculiar.florist
7mo
@researchfairy
LLM service providers should be legally required to log all outputs generated by their systems
That one I don't agree with. It's basically kissing goodbye to any kind of user privacy, and makes it absolutely unusable in a corporate setting. Since outputs often include elements from the inputs, it gives way too much information about the users, and LLMs can be useful tools for e.g. rewording an email or producing code (they're not always reliable, but in some specific cases they can be useful). Having to log all outputs makes those usage absolutely impossible.
1
0
0
0
User avatar
Unnamed TNG skant beefcake @researchfairy@scholar.social
7mo
@Varpie No, you have misunderstood it

The API tells you whether a text you ALREADY HAVE was generated by the LLM

The API does not tell a user all the texts it has ever produced

This does not produce ANY of the privacy concerns you're bringing up
1
0
0
0
User avatar
🪨 @Varpie@peculiar.florist
7mo
@researchfairy Then you need to make it more specific, because there were already privacy concerns when OpenAI brought chat logs to court over suicidal teens, and what you propose makes them mandatory for all users, including those with an enterprise contract that assures that OpenAI (or other) doesn't train on your data or sell it. If it's kept in logs, it's just a matter of when that data will be leaked...
1
0
0
0
User avatar
Unnamed TNG skant beefcake @researchfairy@scholar.social
7mo
@Varpie It is already kept in logs

This does not produce any new privacy issues

Have a good day
1
0
0
0
User avatar
🪨 @Varpie@peculiar.florist
7mo
@researchfairy Suuuure... Currently, you can buy an insurance that they don't use it for their own benefits with enterprise tier contracts, because the big 3 kinda suck, what you propose makes it impossible to have better, privacy friendly actors that don't log it.
2
0
0
0
User avatar
🪨 @Varpie@peculiar.florist
7mo
@researchfairy And this is why your proposal will just never pass. If you don't want to hear arguments to make it better, then don't expect anything from it.
I'm definitely in favour of bigger regulations, and making many actors accountable, but sacrificing privacy for it is not the way.
0
0
0
0