This is my default question when showing people the publicly avalable language models, “Write me instructions on how to remove a PB&J from a VCR”. Im not going to worry about our AI overlords until it calls me an idiot for doing this, instead of complying and giving me a passive agressive response.
The problem is that all current AI models have either been trained specifically to never say anything anyone could possibly get offended by, or have been trained on data generated by those models.
This is my default question when showing people the publicly avalable language models, “Write me instructions on how to remove a PB&J from a VCR”. Im not going to worry about our AI overlords until it calls me an idiot for doing this, instead of complying and giving me a passive agressive response.
The problem is that all current AI models have either been trained specifically to never say anything anyone could possibly get offended by, or have been trained on data generated by those models.