Letter to a prospective student

Introduction

The summer of 2026 has been unforgettable for anyone watching the rapid rise of large language models (LLMs). OpenAI’s release of hundreds of new manuscripts and a breakthrough on the Navier‑Stokes problem have reshaped conversations across mathematics and physics. For undergraduates considering a PhD, especially in theory‑heavy fields like quantum information, the question looms: does the accelerating power of AI diminish the value of a doctoral education? This article reflects on that question, drawing on recent experiences at Caltech and broader community observations.

1. The AI shockwave and undergraduate awareness

During the FUTURE of Physics workshop at Caltech, a group of forty bright, often under‑represented undergraduates arrived eager to learn about PhD pathways. While many asked about funding and lab culture, few mentioned AI at all. This gap mirrors a wider trend: most undergraduates have not yet internalized the speed at which LLMs are entering theorem‑proving and research workflows. The latest models—Fable 5, GPT 5.6, and especially Opus 5.5—have demonstrated capabilities that surprise even seasoned theorists, yet the discourse has largely bypassed the undergraduate audience.

2. How AI is changing the research toolbox

Historically, LLMs were useful for literature searches but unreliable for rigorous proof generation. Over the past year, that landscape has shifted dramatically:

• Theorem‑proving assistance: Models now suggest alternative frameworks, spot hidden assumptions, and even propose plausible conjectures when researchers hit dead ends.
• Benchmark saturation: New theoretical‑physics benchmarks were quickly solved by the latest releases, indicating that AI can handle many routine calculations that once formed the bulk of graduate assessments.
• Question formulation: The hardest part remains translating physical intuition into a well‑posed mathematical problem. AI excels at exploring existing literature, but it still relies on a clear, precise question from the human researcher.

These developments suggest that AI will become a standard part of the research pipeline, much like computational clusters or symbolic software did in previous decades.

3. What a PhD still offers in the AI era

Even as models can generate proofs and draft papers, the core value of a PhD persists:

• Mentorship and judgment: Skilled advisors help students frame the right questions, evaluate AI‑generated results, and navigate the subtleties of scientific rigor.
• Creative freedom: Doctoral training cultivates the ability to pursue high‑risk, long‑term problems that lie beyond the reach of current models.
• Community and leadership: Graduate students learn to write grant proposals, manage collaborations, and contribute to the strategic direction of research programs—roles that remain human‑centric.
• Ethical stewardship: Understanding AI safety and the societal impact of automated research is a responsibility that requires human judgment and interdisciplinary insight.

In short, the PhD experience is less about producing raw calculations and more about developing the judgment, curiosity, and collaborative skills that guide AI tools toward meaningful discovery.

4. Practical advice for prospective students

If you are weighing a theory PhD in 2026, consider the following steps:

1. Seek mentors who embrace AI: Look for advisors who integrate LLMs responsibly into their workflow and can teach you how to critique AI output.
2. Build a strong conceptual foundation: Deep understanding of physics, mathematics, and computational methods will enable you to ask the right questions—something AI cannot replace.
3. Engage with interdisciplinary communities: Workshops like FUTURE expose you to diverse perspectives and help you identify where AI can complement, rather than replace, human insight.
4. Stay informed about AI developments: Follow reputable sources on model releases and benchmark results, but critically assess how they apply to your research interests.
5. Consider the broader impact: Reflect on how your work can contribute to AI safety, scientific literacy, and the responsible deployment of powerful tools.

Closing thoughts

The rapid progress of AI does not signal the end of theoretical physics PhDs; it signals a transformation. The most rewarding path will be one where you learn to harness AI as a collaborative partner, while cultivating the independent thinking and mentorship that only a human research community can provide. By entering a program that values both technical depth and ethical responsibility, you can help shape a future where AI amplifies, rather than eclipses, human scientific inquiry.