Adaptive systems promise Bloom's tutor for every learner. But the same systems compress curricula into whatever can be auto-assessed, and procurement rewards uniformity. Personalisation of pace is not the same as personalisation of purpose.
The tension
Adaptive pacing and curricular narrowing can happen in the same product, at the same time.
Evidence
Each attachment carries its provenance, its method, and the claim it supports or challenges.
Supports Tutoring quality tracks model capability, which is currently a predictable function of compute — so personalisation improves as systems scale.
Provenance Replication of published scaling-law figures, three model sizes · 2024
Method Median loss across held-out evaluation tasks, log-log axes, power-law fits of the form L = AC^-α. Loss is a proxy for capability, not for pedagogy.
Added by @kenji
0 people have plotted themselves
Micro-debates
Thesis, counter-thesis, rebuttal, closing — then the room scores the reasoning.
No micro-debate has been staged on this idea yet.
Each perspective carries a rationale and the evidence behind it.
Standardisation, decisively. Personalisation in current products means pacing through a fixed sequence. The sequence itself is set by what an autograder can mark, which narrows the curriculum toward the machine-checkable.
Both, split by resources. Well-funded schools will use these tools to free teacher time for seminar work; under-funded systems will use them to replace it. The technology is an amplifier of the staffing budget.
Every instructional technology since the lantern slide promised individualisation and delivered administration. I want to know what is structurally different this time before committing to a stance.
Personalised, in the one dimension that has the best evidence behind it: spacing and retrieval scheduled per learner. That alone recovers a meaningful slice of Bloom's two sigma, and no human teacher can do it for thirty students at once.
Not asking whether it works. Asking what specific capability would change your mind, stated in advance.
Research paper
Benjamin S. Bloom · 1984
Reports that tutored students outperformed conventionally taught peers by roughly two standard deviations, framing the challenge of scaling tutoring.
Article
Rosalind Grant · 2025
Field reporting from three school systems piloting adaptive learning software, and the gap between the demo and the classroom.
Book
Nicholas Carr · 2010
Argues that networked media reshapes reading into scanning, trading depth of comprehension for breadth of exposure.