The machines are fine. I'm worried about us

The test case: Alice and Bob

It has been reported that an Ergosphere post lays out a simple, sharp thought experiment: a new assistant professor hires two PhD students, Alice and Bob, gives them comparable, month-sized problems that should take a year for a novice, and watches them work. Alice reads papers with a pencil in hand, stumbles, re-reads, and learns the craft. Bob, allegedly, leans on an AI agent that summarizes papers, explains methods, debugs code, and drafts the manuscript. On the surface? Two papers, two publications, identical output. Under the hood? Two very different graduates.

Why the similarity is dangerous

This is the emotional heart of the piece: metrics mislead. Universities, funders and hiring committees are built to count things — papers, citations, grants — and by those counts Alice and Bob are interchangeable. The worry is not that the machines fail. It's that the system rewards visible outputs while the invisible process — the learning, the intuition, the ability to recover from a sign error at 2 a.m. — is being zipped up and handed to an assistant. What happens when the majority of trainees, many of whom it has been reported will leave academia within a few years, have been honed by proxies rather than practice? Skills atrophy. Tacit knowledge vanishes. The field gets faster but shallower.

What institutions might do (if they care)

This isn't a call to ban helpful tools. It's a prompt to rethink evaluation. Peer review could probe provenance more deeply. PhD programs might restore assessments of process: oral defenses that test lived understanding, open notebooks, reproducibility checks, apprenticeship-style mentorship. Short term? Expect cleaner papers and faster turnarounds. Long term? Ask yourself: do you want a generation that knows how the machine made the music — or only how to press play?

Sources: ergosphere.blog, Lobsters