Skip to content- Sources: primary, discussion
- Summary: The thread responds to a post on Andrej Karpathy's own account arguing that the gap in shared understanding of LLM capability is widening. It drew 647 points and about 940 comments, the largest comment count on Hacker News for this date. It centers on review throughput and names no specific model, product, or API surface, which is why it is carried as pulse rather than as an AI item.
- Why it matters: If review is the binding constraint, the next throughput gain for a team comes from changing how changes get reviewed rather than from upgrading a model.
- Comments: Commenters report that code review at their own employers has become a formality, and argue from that experience that review throughput rather than model capability now sets the limit.