the latest on AI:
1. Language models are optimized for coherence, not truth
The core objective during training is to produce text that is likely and coherent given the context. That objective does not directly encode a concept of factual truth. As a result, the system can produce [garbage]...
3 months ago