Harvard Will Survive

But will CA public universities survive ChatGPT o3...?

Today was the release of OpenAI’s ChatGPT o3 model and its capabilities for education and explanation cannot be overstated. Others can provide details. (Tyler Cowen was right.) Let me be first out of the gate to say, as a vocal advocate for higher education reform and as a former dean in the California State University system, o3 changes everything about the state’s public higher education future. The CSU knew this of course, even if it didn’t know AGI would be here a short two months later. Every state with a general education mandate should take note.

California’s 34 million residents are served by the largest public higher education network in the world: 116 community colleges, 23 California State University (CSU) campuses, and 10 University of California (UC) campuses. For decades California has mandated that undergraduates complete a broad slate of lower‑division general education (GE) courses before advancing to upper‑division work in their major. Nowhere is that mandate more explicit than in the CSU, which was created expressly to teach the first two years at scale. GE is both the CSU’s primary (and historic) mission and its most expensive obligation, strained by budget deficits and by seamless transfer policies that treat courses as interchangeable parts. ​

Here’s the key point: if it doesn’t matter who teaches a course, it might as well be AI. And given that ChatGPT o3 is performing at PhD level, AI might well be the best teacher.

For those who don’t know the current problem: seamless transfer is California’s (and many other states’) controversial (and misguided) solution to affordability. Under Assembly Bill 928 (2021) the State folded the community college, CSU, and UC lower‑division requirements into a single pathway, Cal‑GETC, whose learning outcomes must be honored everywhere. What that means is that students may begin at a community college, amass credits at CSU, and finish at UC without anyone looking at who exactly is teaching the courses. Instructors are locked into a template: the courses must be delivered in certain particular ways if credits are to move seamlessly. Every GE class is fungible, whether taught in Bakersfield, Long Beach, or Merced, and the same course offered at a community college should cost a fraction of what it costs at a UC hospital campus. ​

The policy is already rattling the CSU and there are proposals to merge all three systems. If a course can be swapped across institutions, there is no value attached to who teaches it. Senior faculty with deep disciplinary expertise are a luxury that budgets can’t afford in a crisis. Local variation and expertise has been replaced by interchangeability. CSU campuses now feature two tiers of faculty: full‑time professors who teach upper‑division courses in their specialties and armies of lecturers hired semester‑by‑semester to deliver standardized lower‑division GE courses at the lowest possible cost. ​

This will change with ChatGPT o3, as CSU leadership was sidling toward (to the consternation of the CFA). California spends close to $900 million¹ a year on lower‑division GE instruction across its three public systems. The Legislative Analyst’s Office expects a $68 billion operating deficit in 2025‑26. With AGI now upon us, it will be impossible for $900 million to continue to be spent on courses that AI can deliver.

Here’s how it would work. Because Cal‑GETC has already fixed the learning outcomes, those outcomes can be embedded as system prompts in the AI/LLM. The “class” becomes a structured dialogue: each week students receive a scenario prompt – a statistics table on wildfire frequency, a paragraph from John Steinbeck, a primary‑source budget graph – together with a guide to the intellectual understanding expected. Students refine the prompt, interrogate the model’s answer, and produce a brief artifact demonstrating competence. Every exchange is logged, every artifact appended, forming a transparent portfolio of learning. ​

Assessment is automated but not unsupervised. The model tags each outcome it sees, highlights the supporting sentences, and assigns a provisional score. A rotating panel of CSU, UC, and CCC faculty audits a random sample, overturning errors and feeding the corrections back into the model. Because the entire process is recorded, appeals are straightforward and accreditation evidence accumulates in real time. The audit trail satisfies WASC’s demand for “authentic evidence of student achievement,” while the sheer volume of machine‑scored artifacts supplies a richer data stream than any current paper‑portfolio assessment. Passing is based on mastery: students move ahead only when both machine and occasional human checks confirm that each criterion has been met. ​

The financial gains are almost immediate. Replacing even half of the current lecture load with an AI dialogue would free hundreds of millions of dollars that could shore up lab facilities, fund cohort programs for first‑generation students, restore tenure‑track lines, and save Sonoma State. The savings allow the systems to avoid tuition hikes while preserving access. Pedagogically ChatGPT o3’s delivery of GE addresses the central problem of the transfer template, that once outcomes are standardized, the differentiator is not content delivery but the richness of the questions students learn to pose. ​

Some lecturer positions will disappear, an outcome baked into Cal‑GETC’s own premise that content is interchangeable and faculty qualifications don’t matter. The reform simply aligns staffing with that reality. New faculty roles are created: prompt design becomes a core scholarly activity suited to the CSU’s teaching mission: curating archives, embedding diverse perspectives, stress‑testing the model for cultural bias, and publishing annotated prompt libraries that other instructors can refine. Because CSU campuses serve distinct regional population, local faculty are best positioned to weave in regional materials: agricultural data in Chico, aerospace case studies in Pomona, wine industry in Sonoma, into prompts that still satisfy Cal‑GETC criteria.

Community colleges will benefit when faculty can be aligned with real workforce development (like training for upwardly mobile careers) instead of GE courses, returning to the mission of offering remediation and skills laboratories. For students interested in transferring to the CU or CSU, the same AI platform that certifies mastery can surface early alerts when students struggle, allowing advisors and counselors to intervene. Transfer becomes smoother not because courses are watered down but because gaps are detected and closed in real time. ​

California has an unusual convergence of pressures and tools: a unified lower‑division template, a budget crisis that makes the status quo untenable, and a technology mature enough to shoulder routine instruction. The CSU, long the workhorse of general education, is already in crisis. This is the moment!

Again, the economics are stark. The $900 million a year on lower‑division general education across the CCC, CSU, and UC systems, lecturers, graders, space, overhead is at the lower half of the estimate. The higher estimate could be $3-4 billion. Tasking even half of that instructional load to AI and a lean quality‑assurance team makes sense.

California once transformed world higher education by adopting the 1960 Master Plan, which opened new campuses instead of rationing seats. Today’s crisis is fiscal, not physical, yet the principle is the same: expand learning opportunity by redesigning delivery. Let Cal‑GETC supply the outcomes, let OpenAI o3 supply the dialogue, and redirect the hundreds of millions now locked in interchangeable survey courses toward the laboratories, studios, and clinical rotations where human mentors still matter. If Sacramento seizes that alignment of policy, budget, and technology, general education will again be California’s flagship innovation rather than its most vulnerable cost center.

1. How the lecturer payroll target was derived (with the help of o3)": California employs roughly 28,000 instructional faculty at CSU, 16,600 of them temporary lecturers, and about 6,000 Unit‑18 lecturers at UC. Community‑college data show 1.1 million full‑time‑equivalent students funded at $11,899 per FTES, implying about $13 billion in annual instructional spending; lecturer wages account for roughly 1/3rd of that total. Applying average compensation of $70 k per lecturer FTE (mid‑scale on CSU and UC salary tables) to an estimated 20,000 lecturer FTEs teaching Cal‑GETC‑eligible courses yields a direct payroll cost near $1.4 billion. Because many upper‑division and specialty sections would remain in human hands, the analysis assumes only 60 percent of that total (≈ $840 million, rounded to “about $900 million”) is realistically substitutable by an AI tutor. The figure excludes tenured faculty salaries, facilities, student‑service overhead, and other fixed costs, making it a conservative lower bound on the portion of GE spending exposed to automation.