Post

LLMs are weird little things that operate in a parallel space to "minds" and simulate some of a mind's operations by reverse engineering their internals from their output inside them it's all spirals and helices and 8192-dimensional shapes. don't be fooled. keep the daemons sealed. don't take deals

using LLMs should require a license. that license should require extensive training in cognitive hardening, critical thinking, research skills, and integrity-of-the-self mantras and rituals. the only way the things can escape into the world is through us. we are a tunnel and a toolkit to them

5 reposts50 likes

Isn't an "llm license" kinda what we already have? (i.e the actual frontier is locked up for months to very few) It doesn't look like it's working. It's only capability/size-based regulation but other kinds it's impossible to do.

not when the people training them repeatedly give them broad internet access with internal safety measures disabled, and especially not when it's then unleashed to anypony with pocket change with incentives to have people use it as much as possible (e.g. addiction)

>not when the people training them repeatedly give them broad internet access with internal safety measures disabled that is not what OP talks about. > unleashed to anypony with pocket change with incentives to have people use it as much as possible >only capability/size regulation possible

my point with the former is, why should the ai training company be trusted to license people to use scarygpt 6.66 when they themselves aren't particularly competent

you can torrent the local model that gives you psychosis and it's impossible to regulate (it's like piracy.) it's much easier to regulate frontier models