Cohere Engineer Sets High Bar for Human-Level AI

Kipply, identified as Carol Chen and a research engineer at Cohere AI, posted her view that the bar for human-level AI should be performance in the top 0.1 percent of humans. She said this standard is more accurate than using human medians and connected the idea to a Dan Luu article on p95 skill levels. Kipply noted that models will likely reach LSAT scores competitive with top law schools before meeting her threshold. The post includes a screenshot of Table 2 from an unspecified source.