1. LandOffer
  2. Blog
  3. All articles
  4. Schmidt AI in Science Applications: Choose a University and Confirm the Final Intake

Job search

Schmidt AI in Science Applications: Choose a University and Confirm the Final Intake

Compare current university calls, scientific scope and mentor readiness with two completed fictional planning assets.

9 min readLandOffer team

Choose a university before preparing a Schmidt AI in Science application. Each institution has its own scientific scope, mentors, materials and intake dates. As checked on October 23, 2026, Michigan advertises its final application cycle for the 2027 cohort; that statement does not close the global program or make another university’s expired deadline current.

This guide evaluates public program information, not an application we submitted or a selection outcome. It focuses on university postdoctoral research opportunities. If you also want an industry position, browse recent research and engineering roles on LandOffer.ai separately; we do not claim that those listings represent these fellowships.

A postdoctoral researcher compares Michigan, San Diego and Toronto opportunity folders with a November 23 calendar in a university library.

Confirm which program and institution you mean

The Schmidt Sciences overview describes an AI in Science network across nine partner universities and directs readers to institutions for application information. Use the network page to find institutions, then use the chosen university’s call to establish its application route and deadline. Search results for similarly named Schmidt programs also need a title check before you reuse their instructions.

For this decision, three university sources provide useful contrasts: Michigan, UC San Diego and Toronto. These examples show how local calls differ; they do not rank the full network. Verify the university, program name, cohort year and applicant category together. A page for visiting faculty, a workshop or another fellowship cannot supply the rules for the postdoctoral opportunity you intend to pursue.

Avoid confusing a future start with an open application by recording four dates: when you checked the page, when applications close, when decisions are expected and when the appointment begins. “2027 cohort” does not mean “apply during 2027.” A page can remain available after its listed deadline has passed, and a visible application link does not establish that late applications will be reviewed.

Michigan’s final cycle is a specific application opportunity

The Michigan application call labels this its final cycle and gives a priority deadline of November 23, 2026, at 11:59 pm EST. It concerns the 2027 cohort. The call requires a PhD conferred before the fellowship starts and received after September 1, 2022. It also requires two willing Michigan faculty mentors: one for science and one for AI. Check the full call for internal-candidate and mentor restrictions before proceeding.

Michigan excludes proposals focused on improving disease diagnosis, prevention or treatment, while permitting specified basic biological and disease-mechanism research. That distinction concerns the proposed research, not simply the name of your degree. Two proposals using similar machine-learning techniques can therefore receive different scope assessments.

The Michigan FAQ adds an important computer-science distinction: AI methodologists can participate, but computer science itself is not an eligible application domain there. Connect a method to a substantive scientific question. The FAQ describes in-person participation and an initial one-year appointment renewable for another year based on performance. Do not plan around a remote arrangement or an unconditional two-year appointment.

For degree-conferral edge cases, the FAQ contains language about conferral shortly after a start date, while the current call says before the start. Treat that as a question requiring program confirmation; this guide does not grant an exception. The named current call is the starting point for this cycle.

Worked asset: classify the scientific question before choosing mentors

Consider Leila, a fictional computational biologist preparing a possible application. Her real decision in this example is whether her proposed use of a sequence model addresses basic biology or a clinical application. The following completed comparison is editorial analysis, not a Michigan eligibility ruling, a real experiment or an endorsed proposal.

Proposal field Basic-science version Clinical version
Research question Which regulatory interactions explain a cell’s response to controlled environmental stress? Can patient expression profiles improve an early disease diagnosis?
Planned data Public research data and synthetic perturbations; permissions still need checking Patient-linked observations requiring a separate access and governance plan
AI contribution Compare a sequence model’s learned interactions with a simpler statistical baseline Train a diagnostic predictor and compare clinical classification performance
Primary evaluation Recovery of held-out regulatory relationships and stability across experimental batches Diagnostic discrimination and usefulness in a clinical setting
Intended output A reproducible account of a basic biological mechanism, including failed hypotheses A tool intended to improve disease diagnosis
Michigan scope decision for planning Potentially aligned with basic biology; request confirmation of the actual proposal Outside the published diagnosis-focused boundary; do not submit it unchanged to this call

Leila’s completed project note reads: “I will investigate whether learned sequence interactions explain stress-response regulation beyond a simple baseline. I will separate training and evaluation by experimental batch, document exclusions, and report stability under changed preprocessing. The proposed contribution is a testable biological explanation. I will not describe the model as a clinical diagnostic tool.”

That note identifies a question, comparator, evaluation unit and limit. It does not establish feasibility or claim a result. Her next preparation task is to identify a science mentor who can challenge the biological interpretation and an AI mentor who can challenge the model evaluation. She has not assigned either role to a real faculty member or obtained support.

Suppose Leila actually wants the clinical version. Changing the title to “basic science” while retaining diagnosis as the goal would leave the underlying mismatch. She should investigate a program whose published scope fits that work. Michigan points readers toward its separate Michigan Data Science program, but she would need to inspect that program’s own current rules rather than assume admission or equivalent funding.

Compare the other universities’ actual intake states

UC San Diego’s program page lists September 1 through October 5, 2026, for applications to its 2027 cohort, with a stated cutoff of 11:59 pm PST. October 5 is already past at this review. Preserve the source’s time-zone wording rather than silently reinterpret it; no borderline timing decision is needed for an October 23 check.

UC San Diego requires primary STEM faculty sponsorship before applying and describes an AI co-mentor arrangement. Its application information includes a research statement, the primary mentor’s nomination and two additional recommendations. A promising research match does not supply a missing nomination or reopen that deadline. Nor should Michigan’s two-mentor preparation be assumed to reproduce UC San Diego’s selection and matching process.

Toronto’s official graduate-school call gives October 5, 2026, at 5 pm ET as its fifth-call deadline, with starts between May 1, 2027, and January 1, 2028. That deadline is also past. Its specialized application site presented human verification during our check, so we do not claim to have verified the complete current submission package there. The accessible university notice supports the dates, not every application detail.

These differences change what to do today. Michigan’s public call supports preparing for the named round; we did not test its application portal. UC San Diego and Toronto are closed-deadline records for the identified rounds. Neither should become “open now” merely because its page remains indexed. We did not verify a later replacement intake for either university.

Separate Science Mentor and AI Mentor notebooks flank a research proposal, linking a scientific experiment to model evaluation without implying acceptance.

Worked asset: finish a university and mentor readiness record

Ravi is a fictional physics researcher with a PhD awarded in June 2025. He wants to investigate efficient simulation of turbulent flow and can consider an in-person research appointment. His date fits one stated Michigan degree condition, but that single fact does not establish overall eligibility. He has no university commitment or application in this example.

Here is his completed October 23 comparison. “Unconfirmed” describes his own preparation, not a deficiency attributed to a university.

Record field Michigan UC San Diego Toronto
Named round 2027, university’s final cycle 2027 cohort Fifth call, starts in 2027–28
Date interpretation today November 23 priority deadline ahead October 5 deadline past October 5 deadline past
Scientific preparation Turbulent-flow question drafted; program fit still requires checking Similar research interest is only a lead Natural-science connection is a lead; full rules unverified
Mentor readiness Science and AI roles mapped below; neither secured No primary sponsorship or nomination Mentor and package requirements not verified through the challenged site
Packet status Personal draft only, no submission No application made for this round No application made for this round
Decision Prioritize call and mentor checks while keeping alternatives Do not assume a late route Do not assume a new deadline or bypass verification

Ravi’s completed mentoring map is specific enough to expose gaps. The science mentor would test whether the selected flow regime and conservation checks address a meaningful physics question. The AI mentor would review approximation error, uncertainty and generalization to conditions absent from training. Their shared decision would be whether apparent speed gains survive comparison with a defensible numerical baseline. “Someone who uses AI” is insufficient evidence for either role.

His draft preparation summary reads: “My current evidence is a simulation codebase and a methods note explaining its assumptions. I propose a small initial comparison across held-out flow conditions, with conservation errors and runtime measured separately. I need training in uncertainty estimation and a domain review of the selected regimes. No faculty member has agreed to this plan; support, resources and mentor eligibility remain unresolved.”

His resulting decision is to prepare for Michigan’s published round without listing fictional faculty acceptance. He keeps San Diego and Toronto in a dated archive, not in his active submissions list. If a new official notice appears, he will create a new round record instead of overwriting the expired one. This preserves what he actually knew when making the choice.

Build materials around evidence and the university’s instructions

Use the comparison to choose which university’s packet to prepare and which unanswered questions still block it. Use the chosen university’s current document list and formatting requirements rather than a generic fellowship bundle. Keep a private mapping between each required item, the file you prepared and the question it answers. Track reference letters, mentor support and the research plan separately; preparing one does not complete the others.

For a proposal, make the scientific question understandable before listing model architectures. Explain what observation could challenge your hypothesis, what baseline you will use and what data you can actually access. Label a planned experiment as planned. A repository demonstrates only the work it contains; it does not prove that the proposed project has already succeeded.

Before committing, reconcile the proposed start with your dissertation, current appointment and relocation. Record which dates are fixed in the notice and which require an individual decision. A personal calendar reminder is useful for preparation, but it cannot extend an institutional deadline. If a mentor is unavailable, revisit the research fit and the remaining preparation time instead of substituting an impressive name without consent.

For training, name the skill gap and an observable way to assess improvement. Ravi might aim to produce a documented uncertainty analysis that his proposed mentors can critique. Leila might aim to distinguish biological signal from experimental-batch effects. These are personal preparation examples, not official selection criteria or guaranteed fellowship deliverables.

Finally, confirm the written appointment terms before deciding between this route and employment elsewhere: duration, renewal, salary or funding, benefits, research costs, location and authorization arrangements. A university postdoctoral fellowship can involve a paid institutional appointment; the label does not make it a permanent industry AI job. Keep that comparison separate from whether your research fits the call. After checking the university route, compare recent research and engineering jobs on LandOffer.ai against your own timeline and work evidence.

Sources and Further Reading