Natanael

This gets at my own personal perspective of using LLMs to respond - it’s not just about not putting effort into understanding and responding yourself, rather it is about making yourself a proxy to a tool I could use myself, and doing so *without even having a better understanding of how to use the tool to answer my question*, and still thinking you’re somehow made a positive contribution, that is the most disrespectful.

If you genuinely thought the LLM could help me then you should be explaining your process to me for how to use it and validate responses, or else at least you should ask me for more info and explain how you think it’s responses could help if you really do think you’re better at operating it.

Imagine doing the same in a workshop, and taking a powertool to an object before you even bothered figuring out what the other person wanted. Or trying to be helpful by asking questions on your behalf to other departments, but messing up the context and thus repeatedly producing useless answers that you have to put time into refuting.

Nobody removed your choice to change servers

It’s literally part of the point of the fediverse that you can pick hosts who agree with you, who can apply moderation and bans you agree with on your behalf, and if you disagree you move and bring people with you if you can convince them

And why would a bunch of never-left-their-staters even care in the first place? They clearly don’t want freedom of movement

It’s actually kinda easy. Neural networks are just weirder than usual logic gate circuits. You can program them just the same and insert explicit controlled logic and deterministic behavior. To somebody who don’t know the details of LLM training, they wouldn’t be able to tell much of a difference. It will be packaged as a bundle of node weights and work with the same interfaces and all.

The reason that doesn’t work well if you try to insert strict logic into a traditional LLM despite the node properties being well known is because of how intricately interwoven and mutually dependent all the different parts of the network is (that’s why it’s a LARGE language model). You can’t just arbitrarily edit anything or insert more nodes or replace logic, you don’t know what you might break. It’s easier to place inserted logic outside of the LLM network and train the model to interact with it (“tool use”).

Some of these people are broken enough that they feel they are protecting “their own” by going on the attack against “the others”.

As long as they can still be led to believe somebody they don’t know is dangerous without evidence, then increasing empathy isn’t a solution by itself. You have to either force them to think twice, apply critical thinking, question sources and motives, demand evidence, etc, or else simply make them stop.

I would force it on the hysterical right wingers who believe they need to impose their will on the world

Then people would actually be happy

Use it as a dumb worklog sometimes. Tell it what you were already gonna do and ask if it would do the same. It’s almost always gonna agree. Then just ignore it. If somebody AI obsessed pulls out the full logs they’re gonna see you’re doing what the AI said was good. (basically Inception, lol)

I’ve heard of that happening in context of thieves breaking into stores. Never heard of it used for home robbery