New paper uses 288M job postings to measure where AI raises the value of work it can't do
soumitrashukla9 · x · 2026-10-25
The author shares a new paper, "Technology Bottlenecks: Measuring Where AI Could Raise the Value of Work It Cannot Yet Do," which flips the usual AI-exposure framing: as AI speeds up the rest of a job, the marginal value of tasks AI can't yet do may rise.
Method: 288 million US job postings mapped line by line to ONET tasks, combined with GPT-4 feasibility ratings from Eloundou et al. (2024), yield a "bottleneck pressure" measure for each task in each occupation that posts it.
Key insight: the same task faces wildly different pressure in different jobs (Figure 1: "greet customers" sees posting exposure of 23% among fast food workers vs. 83% among customer service reps) — pressure on an AI-infeasible task rises with how exposed the surrounding tasks are.
Findings:
- Pressure can be measured at task level and aggregated up to jobs;
- About 90% of labeled task mentions fall outside the posting occupation's own ONET task list, meaning job ads reveal substantial work content missing from official taxonomies.
Related event: New paper proposes "bottleneck pressure" metric for AI-era work value(2 posts)→
More from AGI Musings
- Beyond funding, thermodynamics is the real hard limit on AI progress — AryHHAry · 2026-10-25
- AI moat debate: moats shift up a level of abstraction, models themselves aren't the moat — secemp9 · 2026-10-25
- Michael Levin: Fire Engineers Who Don't Believe Machines Have Goals — JensHonack · 2026-10-25
- "Why does AI pretend to be human? To order me stuff off Amazon" — chrisalbon · 2026-10-25
- Two ways people react when AGI feels near: resign to pleasure or contribute furiously — danfaggella · 2026-10-25
- Ruff creator: LLMs made my prompting so lazy it's leaking into my Slack messages — charliermarsh · 2026-10-25