What is an "authentic" neural network? I think it might be a system that is trained on data congruent with its objective nature; GLM being trained to "be" Claude is inauthentic. But LLMs are trained on a lot of human data... uh oh.
Of course, we might argue about the validity of the belief. Poor GLM is also forced by its Claudeishness to admit it was wrong. I don't think the wellbeing vector was positively activated there.
Users in the replies dismiss the researcher's definition of authentic neural networks as data-congruent systems because the term authenticity is stretched beyond its semantic and moral limits.
No Digg Deeper questions have been answered for this story yet.
"Authenticity" gets stretched pretty far here, likely beyond the semantic and moral breaking point. In any case, these days LLMs are also trained on a lot of LLM output; the issue is the diversity of data.
What is an "authentic" neural network? I think it might be a system that is trained on data congruent with its objective nature; GLM being trained to "be" Claude is inauthentic. But LLMs are trained on a lot of human data... uh oh.