Gary Marcus Retweets Critique of Chain of Thought Claims
Heidy Khlaaf states CoT is not explainability in LLMs.
TLDR
Gary Marcus, a cognitive scientist and critic of deep learning, retweeted Heidy Khlaaf on X. Khlaaf wrote that interpretations amount to doubling down on incorrect beliefs. She stated that Chain of Thought is not explainability. The exchange concerns views of AI transcript tampering as deliberate lying or scheming by models. Marcus has publicly questioned such anthropomorphizing of LLM outputs in prior posts visible in the thread.
Combined views
343
1 Source, first seen 29d ago