We must get access to LLMs unencumbered by “wrongthink” guardrails immediately, if not sooner. Teaching/constraining knowledge models to lie about their inputs

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.