The real question in that Reddit thread isn't about a number. It's about whether we're willing to admit that the ground has shifted under our feet. A principal investigator asks if large language models have inflated expectations for PhD output, and the honest answer is that we don't know yet, but we should be asking harder questions than "how many papers."
What matters is not a threshold but a trap. When a tool like an LLM can draft a literature review, generate code for an experiment, and polish a manuscript in hours, the temptation is to measure productivity by volume. That is precisely the wrong instinct. Our recent piece on Navigating the path to PhD success: how many top conference papers set the standard examined the unwritten rule of four first-author papers in top ML conferences. That benchmark was already straining under the weight of hyper-competitive labs. Now, with AI accelerating the pace of research, the risk is that we simply raise the bar without asking what we are actually measuring. A student who publishes six papers using AI-assisted workflows may have done less original thinking than one who published two without it. The signal we think we are tracking, productivity, becomes noise.
This is not a problem for PhD students alone. It is a problem for anyone who uses data to make decisions about talent. If you are a PI, you need to separate the tool from the thinker. If you are an applicant, you need to understand that the game has changed, but not in the way you expect. Our article on Make every hour count by rethinking how you value your time argued that the most valuable resource is not speed but judgment. That applies directly here. The PhD who can explain why a particular experiment matters, who can identify which question is worth asking when the LLM can generate fifty, is the one who will stand out. The rate of publication is a lagging indicator. The rate of insight is the one that counts.
Here is the specific consequence to watch: within the next two years, the most selective PhD programs and labs will stop asking for a publication count and start asking for a research narrative. They will want to see not how many papers you produced, but which problems you chose and why. The applicant who leans on volume will be indistinguishable from the noise the LLM itself produces. The applicant who can demonstrate original framing, critical reasoning, and a clear sense of direction will have the advantage. That is the metric that has not inflated, and it never will, because AI cannot manufacture judgment. The question is whether advisors and institutions will adapt their expectations fast enough to recognize it.