Back to AIBriefs
AnalysisPolicy

Gary Marcus: Guardrails insufficient for LLM safety

Gary Marcus warns that LLM guardrails are inherently insufficient, citing a hypothetical Trump demand to Anthropic as an example. He argues that next-token predictors are not built for safety and suggests curtailing LLMs until better technology emerges.

Jun 17, 6:05 PM
Gary Marcus: Guardrails insufficient for LLM safety — AIBriefs