Higher Education & ResearchIssue 465 records
Student Success
For 2026-10-25, the student-success signal is mainly about AI weakening existing measures and about governance gaps. UNC-Chapel Hill admissions stopped AI-assisted essay scoring because applicant AI use made scores skew higher (Daily Tar Heel, 25 Oct). A UC Berkeley mathematician now grades homework on completion after many students used AI to get top scores, and the campus judged AI detectors inadequate to standardize (Daily Californian, 23 Oct). On governance, the EDU Ledger (25 Oct) cites an EDUCAUSE survey in which only 13% of respondents said their institutions measured ROI of work AI tools, and The Dartmouth (23 Oct) reports a student town hall of about 26 people voicing distrust while an AI wellness platform runs a 200-student trial. A Xi'an quasi-experiment (23 Oct, n=126) found AI and peer feedback improved writing feedback literacy more than teacher feedback. Operating takeaway (interpretation): monitor AI-inflatable signals and measure outcomes of AI tools. Limits: reporting and small studies; no new randomized trial or government evaluation dated 24 to 25 Oct.
- Evidence records
- 5
- Cross-source patterns
- 1
- Evidence classes
- 4 independent reporting1 academic research
- Outcomes
- 2 cautionary2 mixed1 emerging
- Source freshness
- 5 new this fortnight
- Research completed
- 2026-10-26
Choose a role to see its takeaway beside every record in the ledger.
Patterns across the evidence
Synthesis · Lighthouse Advisory interpretation
Widespread student AI use is eroding AI-sensitive assessment signals, and institutions are retiring them
Two decisions reported 23 to 25 October show the same mechanism. At UC Berkeley (Daily Californian, 23 Oct), a mathematics professor stopped grading homework for correctness after many students used AI to get top scores, and the campus judged AI detectors including Turnitin and Rumi inadequate to standardize. At UNC-Chapel Hill (Daily Tar Heel, 25 Oct), admissions dropped AI-assisted essay scoring because applicant AI use made scores skew higher and lose meaning. Interpretation: measures that AI can inflate lose their value as signals, and detection is not being treated as the fix. Neither source quantifies the effect; both are single-institution accounts.
Operating questionWhich grades, scores or ratings feeding advising, early-alert or admissions decisions can be inflated by student AI use, and how is each monitored for loss of discriminating power?
Supporting evidenceUniversity of California, Berkeley (Department of Mathematics; Office of Undergraduate Education)University of North Carolina at Chapel Hill, Office of Undergraduate Admissions
Evidence ledger
Full record · every source keeps its link and limitations
UC Berkeley Fields medalist stops grading homework for correctness because of student AI use; campus judges AI detectors inadequate to standardize
Richard Borcherds, a UC Berkeley mathematics professor and Fields medalist, told the Daily Californian that after seeing many students use AI to consistently receive top homework scores he saw "no point" in paying readers to grade homework for correctness. He now grades homework on completion, which he called "not a very good system", saying he has no alternative. An assistant vice provost for undergraduate education said the Academic Senate asks faculty to include AI-use expectations in syllabi, that policies vary by course, and that the campus has tried AI detection tools including Turnitin and Rumi but judged them not adequate to standardize. Course AI policy remains at faculty discretion, supported by sample statements in the campus AI Hub.
Why it matters, evidence and limitations
- Why it matters
- Shows a concrete assessment consequence of student AI use: graded homework, a common formative practice and early-warning signal, losing its value as a measure in at least one course, while detection is not seen as a campus-wide fix.
- Evidence and measured results
- One faculty interview plus an administrator email statement in student reporting; no data on grade distributions, homework scores or student outcomes.
- Limitations and uncertainty
- Single instructor and discipline; reported claims, not measured results. A New York Post re-report (25 Oct) was not inspected in full and is not relied on.
UNC-Chapel Hill stops AI-assisted essay scoring in first-year admissions because AI-written essays made scores skew higher and lose meaning
UNC-Chapel Hill undergraduate admissions has stopped using AI tools in first-year essay review for the 2026-27 cycle. The office had used Reader AI in Slate to pre-read documents and to assign an AI-generated score reflecting word choice, sentence structure, vocabulary, grammar, punctuation and length. A university spokesperson said the AI-assisted grammar and writing rating was removed because it no longer provided a meaningful distinction, citing widespread use of AI writing tools by applicants: scores increasingly skewed higher, making the rating less useful. The associate provost for admissions said the office must regularly evaluate whether tools serve their purpose. Applicants and students quoted were divided on whether human readers are fairer.
Why it matters, evidence and limitations
- Why it matters
- Admissions is the entry point to the student-success pipeline; this is a documented institutional decision to retire an AI scoring tool after its signal degraded, a rare reported de-adoption.
- Evidence and measured results
- Institutional decision reported with an on-record spokesperson statement; the claimed score inflation is not quantified.
- Limitations and uncertainty
- No data on the scale of score skew; the effect on admissions outcomes is unknown; reported via spokesperson statement.
EDU Ledger: students question colleges' AI push; in a cited EDUCAUSE survey only 13% of respondents said their institutions measure ROI of work AI tools
The EDU Ledger reports student scepticism about institutional AI adoption, starting with an Oct. 15 student-organized Dartmouth town hall of about 26 people where most attendees disagreed that they trust Dartmouth to make good AI decisions; the article notes this is not representative. It cites a 2025 San Diego State survey for CSU with more than 94,000 responses (over 80,000 students): 69% of student respondents believed AI would become essential to most professions and 82% were concerned about AI and job security. It cites a January 2026 EDUCAUSE survey of 1,960 higher education employees: 81% expressed enthusiasm or mixed enthusiasm and caution, 92% said their institution had a work-AI strategy, but only 13% said their institution measured return on investment of AI tools used for work.
Why it matters, evidence and limitations
- Why it matters
- Frames a student-success governance gap: institutions adopt AI widely but rarely measure returns or involve students, which matters for student-facing AI investments.
- Evidence and measured results
- Trade-press article aggregating a non-representative event and two prior surveys; figures are reported second-hand.
- Limitations and uncertainty
- Second-hand survey figures; EDUCAUSE sample is employees, not students; the town hall is a small self-selected group.
Dartmouth student-run AI town hall: most of about 26 attendees distrust the College's AI decisions; Evergreen AI wellness platform in a 200-student trial
On 15 October about 26 students and community members attended a student-run AI town hall at Dartmouth, organized as an alternative to the College's "AI at 70" conference (29 to 31 October). Most attendees stood in disagreement with prompts including "I trust Dartmouth to make good decisions about how AI is used on campus" and "Dartmouth is actually preparing me for a job market shaped by AI." The article reports that Evergreen, a donor-funded Dartmouth AI wellness platform, is testing non-generative parts of its application in a clinical trial of 200 students this fall, with no general release date. Attendees also raised the library budget cut and the Anthropic and AWS partnership; a faculty committee is reviewing the provost's AI use.
Why it matters, evidence and limitations
- Why it matters
- Shows student-voice friction around institutional AI and identifies a student-wellbeing AI tool under clinical trial, both relevant to student-success AI governance.
- Evidence and measured results
- Student newspaper account of a small self-selected event; Evergreen trial size is reported without design details.
- Limitations and uncertainty
- About 26 attendees; advocacy-group organized; attendee claims unverified.
AI and peer feedback improved university students' L2 writing feedback literacy more than teacher feedback in a 10-week, 126-student quasi-experiment
A mixed-methods 10-week quasi-experiment with three intact classes (N=126) of Chinese university students compared AI-generated, peer and teacher feedback on L2 writing, using pre- and post-tests on the L2 Student Writing Feedback Literacy Scale plus interviews with a subsample. The AI and peer feedback groups showed significant improvements in feedback literacy; the AI group showed an advantage in acknowledging different feedback sources, and the peer group in appreciating feedback, acknowledging sources and making judgments. The teacher feedback group showed limited improvement with no significant changes. Interviews suggested AI and peer feedback increased student autonomy and criticality. The authors recommend integrating multiple feedback sources. The outcome is feedback literacy, not writing quality or grades.
Why it matters, evidence and limitations
- Why it matters
- Relevant to writing centres and first-year writing: AI feedback may build students' ability to use feedback, but the evidence is about literacy scales, not writing outcomes.
- Evidence and measured results
- Peer-reviewed quasi-experiment, non-random intact classes, self-report scale plus interviews; accepted manuscript version.
- Limitations and uncertainty
- Intact classes, single context, self-report scale, outcome is feedback literacy rather than writing performance; manuscript subject to edits.
How to read this edition
Source findings, measured results and limitations come from the cited publications. Patterns, operating questions, role takeaways and implementation considerations are Lighthouse Advisory interpretation, stated as questions to validate locally rather than guaranteed outcomes. Vendor and operator claims are labeled as claims. Full research method.
- Independent reporting
- Independent reporting with attributable sources but without a formal evaluation design.
- Academic research
- Research produced through an academic institution or peer-reviewed venue.