Daniel Lemire: AI now writes passable PhD theses, but despair is the wrong response
Daniel Lemire, writing on Twitter/X, opens by quoting Tony Feng, a mathematics professor at UC Berkeley: in the past, earning a math PhD built resilience, resourcefulness, critical thinking and healthy skepticism, but now a passable PhD thesis can be produced with the push of a button. Lemire says the same holds outside math, in sociology, education, computer science or mechanical engineering.
He traces how research tools changed across his own career. Before the web, research meant physical libraries, paper journals, and photocopies; graduates finished with personal stashes of articles they had to memorize since there was no search function. Then came the web and keyword search over dumped PDFs, then Google Scholar, which let a lone researcher with library access survey a field's literature in days rather than years. Meanwhile the volume of PhDs kept climbing: US PhDs awarded rose from about 36,000 in 1990 to more than 58,000 in 2024, while China's grew from a few thousand a year in the early 1990s to roughly 88,000 a year today. Now agentic AI is, in Lemire's telling, the biggest jump yet: he says he could build a PhD thesis generator today that takes a topic and outputs a PDF resembling what would have passed as a PhD thesis in 2015, adding that most theses were never good writing to begin with, so the bar to clear is lower than people assume.
On the usual objection that a live defense and oral questions still catch weak candidates, Lemire counters that failing a defense is, for social reasons, uncommon; some candidates are simply bad at thinking on their feet and get confused, but that rarely sinks them. He recalls attending a defense recently where an older colleague, who says he opposes ChatGPT and similar tools, kept praising how a young candidate had produced an unusually extensive literature review in very little time. Lemire says nothing, unwilling to be rude on purpose.
Summing up the sentiment he has heard from several mathematicians, Lemire writes that the foundation of modern academia has been blown to bits, not shaken but wiped out. He then breaks from that camp: he calls the sentiment understandable but dangerously wrong, arguing mathematicians need to adapt hard and fast rather than hope the disruption from AI will go away. The post drew 42 points and 37 comments on Hacker News, where some commenters described watching students outsource entire analyses to ChatGPT, while others argued a PhD's real value has always been the training it gives the student rather than the thesis document, and that academia's dysfunction (including retaliation against people who flag suspected research fraud) predates AI.
Key facts
- Tony Feng, a UC Berkeley mathematics professor, told Lemire that a math PhD used to build resilience and critical thinking, but 'suddenly you can produce a passable Ph.D. thesis with the push of a button.'
- Lemire says agentic AI could today build a PhD thesis generator that takes a topic and produces a PDF resembling one that would have passed in 2015.
- US PhDs awarded rose from about 36,000 in 1990 to over 58,000 in 2024; China's rose from a few thousand a year in the early 1990s to roughly 88,000 a year today.
- Lemire recounts a colleague praising a candidate's unusually fast, extensive literature review at a recent oral defense, without realizing it may have been AI-assisted; failing a defense is rare for social reasons.
- Despite summarizing the view that academia's foundation has been 'blown to bits, not shaken, wiped out,' Lemire calls that verdict dangerously wrong and says mathematicians need to adapt fast rather than wait for the disruption to pass.
Why it matters
The essay frames agentic AI as the latest step in a decades-long shift in how research gets done: from physical libraries, to keyword search over PDFs, to Google Scholar, to tools capable of producing a document resembling a passable PhD thesis on demand. That threatens the implicit assumption behind a PhD, that the credential reflects years of resilience and independent critical thinking, since the artifact that used to prove that can now be generated quickly.
Who it affects
PhD students and their supervisors across fields Lemire names beyond math, including sociology, education, computer science and mechanical engineering; universities that run oral defenses as the remaining check on thesis quality; and, by Lemire's account, a wider group of mathematicians who share Tony Feng's alarm that the foundation of academia has been wiped out.
How to use it
The piece offers no detection tool or policy fix. Lemire's practical takeaway is a call to adapt: instead of trying to make the disruption go away, mathematicians and academics generally need to change fast. He implies oral defenses are not a reliable backstop, since failing one is socially rare even when a candidate cannot think on their feet, but he does not propose a replacement mechanism.
How solid is it
The claims rest on Lemire's personal career history, a secondhand quote from Tony Feng, and one anecdote about a colleague praising a literature review at a defense. The PhD-count figures for the US and China are stated without a cited source in the text. The piece drew a substantial Hacker News discussion (42 points, 37 comments), but that reflects engagement with the argument, not independent verification of its more sweeping claims.
Risks and caveats
The central anecdote is a single, secondhand data point, not evidence of a broad pattern. Commenters pushed back on Hacker News, arguing a PhD's real value has always been the training it gives a student rather than the thesis document, and that some of academia's dysfunction, such as retaliation against people flagging suspected research fraud, predates AI. Lemire's own conclusion also cuts against the alarmist framing quoted in the headline: he explicitly calls the 'foundation blown to bits' verdict dangerously wrong.
“In the past, the path to a math Ph.D. cultivated resilience, resourcefulness, critical thinking, and a healthy skepticism, but suddenly you can produce a passable Ph.D. thesis with the push of a button.”
— Tony Feng, UC Berkeley mathematics professor, quoted by Daniel Lemire