The question lands in every CS department group chat at least once a semester: if an LLM can draft a related-work section, generate baseline plots, and refactor a training loop, why are we still treating a PhD like a five-year marathon? Whether these tools are finally compressing the timeline, the honest answer, so far, is no. That is not a failure of the technology. It is a misunderstanding of what a PhD actually measures. The degree was never a reward for producing a certain volume of tokens or charts. It is a certification that you can identify a meaningful problem, design a rigorous investigation, and defend your reasoning under pressure. A model that writes faster does not teach you which experiment is worth running, or why a negative result still matters. The bottleneck has shifted, but it has not disappeared. It has moved from the mechanical labor of typing and coding to the harder, slower work of judgment, taste, and intellectual ownership.
We see this clearly in the gap between "running experiments" and "completing a dissertation." LLMs excel at the former, especially when the task is well-specified. They can scaffold a baseline, suggest a loss function, or summarize ten papers into a coherent narrative. But a PhD committee is not evaluating your ability to produce a polished PDF. They are evaluating your ability to situate that work within a research community, to articulate what your contribution changes about how we think, and to anticipate objections that no model has seen in its training data. Those skills require iterative, messy, often boring human practice. The student who outsources their literature review to a model might save two weeks, but they lose the accidental discoveries that come from reading a paper sideways, following a citation trail, or misremembering a result and having to correct course. That friction is not inefficiency. It is the mechanism by which researchers develop instincts.
For the reader who is considering a fast track, our take is direct: use the tools, but do not mistake acceleration for depth. The most productive use of an LLM in a PhD is not to write your paper for you. It is to act as a tireless, non-judgmental collaborator that can help you explore a hypothesis space more broadly, draft a proof-of-concept that you then scrutinize line by line, or produce a first draft that you can ruthlessly edit into your own voice. The students who finish sooner will not be the ones who automate the writing. They will be the ones who use automation to free up time for the parts of research that only a human can do: choosing the right problem, framing it for a skeptical audience, and building the resilience to keep going when the results are ambiguous. If you are in that position, ask yourself what you would do with the extra two hours a day. If your answer is "write more," you are missing the point. If your answer is "think harder about the next step," you are on the right track.
The open question we are watching is not whether graduation times will drop, but whether the nature of the dissertation itself will evolve to reflect a new division of labor. If a model can handle the mechanics, what remains for the candidate to prove? The answer, we suspect, is a sharper emphasis on research vision and impact, which is a higher bar, not a lower one. So before you celebrate the end of the five-year grind, consider this: the tools may make the first draft easier, but they also make the final defense more revealing. The speed of production was never the constraint. The clarity of thought was always the bottleneck, and it still is.