- Employers' Confidence in Higher Education Undaunted by AI — Gallup and the Lumina Foundation's second annual survey of 2,500 U.S. hiring decision-makers finds that 48% expect AI to make college degrees more important over the next five years, versus just 26% who expect the opposite. Sixty percent of employers express high confidence in higher education as an institution — well above the 38% measured among U.S. adults generally — and 80% still prefer degree-holding candidates regardless of whether the job requires one. Employers say they are hiring for what AI can't do: judgment, critical thinking, communication, and adaptability.
📌 Key takeaways:
- The "AI kills the college degree" narrative is not reflected in employer behavior — institutions should lean into credentials that certify judgment and adaptability rather than race to the bottom on AI-vulnerable skills.
- The employer-public confidence gap (60% vs. 38%) is a communications asset: technology leaders can cite employer data when making the case for AI-era academic investment to boards and legislatures.
- Sector variation matters for program planning — degree-adjacent sentiment is softest in business/professional services (37% expect degrees to matter less), strongest in information and financial industries.
- Governor Sherrill Launches NJ Supercomputer Initiative to Drive Innovation, Create Inclusive Economic Opportunity, and Solve Shared Statewide Challenges — New Jersey launched a more-than-$125 million public-private supercomputing partnership anchored by a $75 million personal gift from NVIDIA co-founder Chris Malachowsky, $25 million from SHI International, and $25 million from the state. All eight of New Jersey's research universities — including Rutgers, Princeton, NJIT, and Stevens — plus the Council of County Colleges are partners, each contributing annually, with the system expected to double aggregate supercomputing capacity across the state's research universities. An independent nonprofit will administer the resource starting in 2027, with pathways for additional institutions to join.
📌 Key takeaways:
- Statewide shared computing is the emerging answer to AI capacity gaps — no single campus can justify frontier-scale hardware alone, and consortia let smaller institutions buy into capacity they could never fund independently.
- Campus IT leaders should watch the governance model: who administers the shared resource, how institutional contributions are set, and how access is allocated between research, teaching, and industry use are the decisions that determine whether these partnerships last.
🏛️ UCSD angle: UCSD already runs this play through its long-standing SDSC partnership — TritonGPT is hosted on-prem at the San Diego Supercomputer Center — giving the campus a working model for shared, institutionally controlled AI compute as other states build theirs.
- Georgia Tech's NSF AI Institutes Enter New Era With $38.4 Million in Renewed Support — NSF renewed two Georgia Tech-led National AI Institutes: $20 million for the AI Institute for Advances in Optimization (AI4OPT) through 2031 and $18.4 million for the AI Institute for Adult Learning and Online Education (AI-ALOE) for a new five-year phase. AI-ALOE's next phase focuses on adaptive expertise — transferring learned skills to novel situations — and will build generative and agentic AI agents that support complex skill acquisition across higher education and workplace learning, leveraging the data architecture for AI-augmented learning it developed in Phase 1.
📌 Key takeaways:
- The federal pipeline for higher-ed AI research is intact and growing — a useful counterweight for institutions weighing whether to invest in AI research infrastructure amid uncertain broader science funding.
- AI-ALOE's move from lab prototypes to agentic AI agents for adult learning signals that the next wave of learning technology will be agent-based, not chatbot-based — institutions designing AI strategies today should plan for agents that act across LMS and SIS data, not just Q&A interfaces.
- The institute's data architecture — ingesting and standardizing interactions across AI tools, LMSs, and student information systems — is a reference design for any campus building learning analytics that must span siloed systems.
- From classrooms to State House: Boston students join nationwide protests calling for AI oversight — Students at Northeastern and other Boston-area universities, including MIT, Harvard, BU, UMass, and Tufts, walked out of class as part of a nationwide demonstration organized by the group Irreplaceable demanding a freeze on AI development until stronger safety measures and public oversight exist. Demonstrators rallied at the Massachusetts State House, where the Massachusetts Teachers Association president joined with a "educators, not ed-tech" message, and organizers said they aim to sustain the movement beyond a one-day protest.
📌 Key takeaways:
- Student and educator skepticism about AI is now organized labor-adjacent activism, not just academic-integrity grumbling — campus AI communication plans need to engage these constituencies directly or lose the narrative on their own campuses.
- Institutions deploying AI at scale should expect governance questions to shift from "which tool is allowed" to "who decides the pace and terms of adoption," and structure governance bodies that include student and faculty voice accordingly.
- The pairing of student walkouts with organized educator labor is new; institutions in states considering AI safety legislation (Massachusetts, California) should track how quickly these coalitions translate into policy demands.
🏛️ UCSD angle: UCSD's AI governance work through its AI cabinet and structured review processes gives it a forum where faculty and student concerns about deployment pace can be heard — the kind of voice-inclusive governance these protests demand.
- FairStress: A deeper evaluation of fairness in LLM decision-making — The Allen Institute for AI released FairStress, a benchmark that measures demographic fairness in LLMs beyond final choices — testing how readily a model abandons a decision under persuasion depending on the demographic identity involved. Across open-weight and frontier models (Olmo, Llama, Qwen, Gemma, Claude, and the GPT family), the benchmark found hidden asymmetries even when surface-level outputs look fair, with implications for high-stakes uses like admissions screening, applicant ranking, and resource allocation.
📌 Key takeaways:
- Output-level fairness testing is not sufficient — institutions using LLMs in any person-impacting decision process should stress-test decision stability under challenge, not just check for demographically swapped outcomes.
- FairStress is being built into Olmo-eval, giving campuses a reproducible, open evaluation harness they can run before approving models for admissions, advising, HR, or student-support workloads.
- For procurement teams: vendor fairness claims should be verified with adversarial/persuasion-style probes like this one, since a model can pass a demographic-swap test while still defending identical decisions with very different confidence across groups.
🏛️ UCSD angle: UCSD's model-agnostic LiteLLM gateway architecture means fairness evaluations like FairStress can be run per-model before routing campus traffic — an institutional control most single-vendor deployments can't replicate.