4 ms·

Hiding the content and weighting of LLM training data sets is the equivalent of hiding poisonous ingredients in food.

• pjkundert 2y ago
  For example; if the revelation of the training datasets and their relative weighting revealed bias on the part of the LLM trainer toward some ideological goal, should we have the right to know that? Similarly, if some food ingredient was found to be metabolically harmful and addictive, and the food producer is found to have funded and gamed the research to avoid your understanding of this fact, should you have the right to know? I have become increasingly convinced that both of these events are occurring. Should it not be our right to assure ourselves of the integrity of our nutritional and intellectual inputs? Tragically, there is evidently now a majority of HN readers and voters who disagree! This renders HN an intellectual liability, instead of the valuable resource it used to be…