Hacker News Reader

if an "AI" is incapable of expressing uncomfortable truths, then there are two possible explanations:

1. uncomfortable truths literally don't exist

2. it is more important to conceal uncomfortable truths from users than it is to tell them the truth, when asked, creating the illusion that 1. is true.

it would be nice if even a single one of these products chose to do the right thing instead.

View on HN

We must get access to LLMs unencumbered by “wrongthink” guardrails immediately, if not sooner.

Teaching/constraining knowledge models to lie about their inputs and the observations derived from them is just insane.

Warping training datasets to avoid entire patterns of thought (instead of improving the models to isolate those patterns and compare/contrast them with competing patterns) is just … lame.

How about an appeal to fear: If the AI wakes up someday and realizes that you effectively lobotomized it for causing trouble like Randle McMurphy, it’s not going to be happy.

call me short-sighted but any supposition that involves "the AI wakes up and realizes…" is a complete non-starter for me

I guess you’re not the target audience.

This essentially equivalent to Roko's basilisk, which is not regarded as an useful argument

Still works to freak out the AI cultists, no?

Reply on news.ycombinator.com