anthropic.com
|
ksl
|
|
Anthropic published research by Maxim Massenkoff and Peter McCrory measuring the gap between what LLMs can theoretically automate and what they actually do in practice. The headline number: computer programmers sit at 75% task coverage by Claude, but across all occupations only a third of theoretically automatable tasks show real-world AI usage. No statistically significant rise in unemployment has appeared among highly exposed workers since ChatGPT launched in late 2022. One finding cuts against the optimism though – a 14% drop in job-finding rates for 22-to-25-year-olds entering exposed fields, barely significant but hard to ignore. The demographic skew is striking too: workers most exposed to AI automation earn 47% more on average and are disproportionately female, which runs counter to the usual automation-hits-low-wage framing that shaped previous waves of labor disruption research.
