reading matters

Lec 2: reading matters

A. issues with trusting AI answers: correctness and difficulty

ask the same pinned LLM a textbook question, then one just past the worked-out literature. Both answers sound fluent, only the first can be checked, and nothing in the response marks the difference
first tempting fix, ask the LLM how confident it is. It gives you a number that means little on its own: calibrated confidence (2023) takes targeted prompting or post-hoc scaling, and detecting confabulation (2024) takes several samples clustered by meaning
second tempting fix, reasoning models. The trace feels like it shows why the answer is right, but it may rationalize after the fact (1977) instead of reporting the computation. Think-aloud protocols (1980) are the ancestor, except that there the person thinking did the talking; here one system writes both the steps and the answer, so reading it is not verifying it
this disconnect is measured: meaningless filler tokens let transformers solve tasks they fail when answering head-on (2024), and models trained on corrupted traces whose steps bear no relation to the problem (2025) perform like models trained on correct ones. A persuasive trace is not evidence
third tempting fix, ask a second provider. Disagreement is useful; agreement is weak, since shared training data, architecture, and fluency incentives correlate their errors (algorithmic monoculture, 2021), and when measured on mistakes rather than answers, model errors overlap well beyond chance and more so as capability rises (ICML 2025). Independence is a property that needs arguing for, not assuming; auditing a model that judges a model is W10’s topic
even assuming LLMs always give a technically correct answer, can we safely just delegate?
how hard a question feels says little about how checkable the answer is; a hard-feeling textbook derivation can be checked step by step, while an easy-feeling question about one’s own project may have no source to check against
nor is difficulty visible in the request itself. A team ran an LLM router for four months and withdrew it (2026), since what makes a task hard shows up later, in what the tool calls turn out to require
so the delegation call stays yours even on optimistic assumptions; nothing along the tech stack (the mechanisms of objective, context, and sampling, covered in W8, the machinery beyond vanilla transformers) signals it

B. the same two questions at the frontier

the systems from W1 return: Co-Scientist (Nature, 2026) is models checking models, generation agents feeding a ranking tournament with a reflection agent as reviewer, running lit review, generation, and ranking at scale, while validity still rests on evaluation and selection outside the loop. the OpenAI field report (2026) found the same pattern across eight real projects: agents handled well-scoped requests with the difficulty sorting already done by humans, the validity call stayed with the researchers, and their time moved from writing code to defining how correctness would be measured, now more managers than engineers
past the literature and past checkable are different boundaries. A counterexample to a flow conjecture open since 1999 came out of a four-prompt chat (2026), and the object is eight routings on seven nodes, so anyone can settle it in an afternoon; an independent verifier did that the next day while the trackers still list it unreviewed, since there’s no paper and no referee.
Checking against credible sources is one way to verify, not the only one; here the check came from understanding the question well enough to run it yourself, a small case of W3’s claim that building grows what you can check
what that case does not prove: the conjecture was a for-all claim, which one object can kill; most questions, including the ones a UROP or MEng project poses, ask how much or whether something works, and under what conditions, questions vaguer, less quantifiable, less mechanical; no single object decides those. The flow example also says nothing about the prompts that returned plausible non-counterexamples.
indeed, the opposite failure exists too: the first claimed LLM solution to an Erdős problem turned out to have been solved by Erdős himself in 1977 (2026), nobody had checked the literature.

C. issues with AI-assisted reading

parts A and B asked what you can tell about the model’s answers, correctness and difficulty; this part asks what the reading was doing for you all along, and what outsourcing it costs
in a controlled CS1 experiment (2024), students with an assistant stopped looking things up elsewhere and learned no more than those without. Agreeing with a summary is the cheapest of the three acts, agree, verify, read, and two results say why that costs you: the generation effect (1978), material you produce is remembered better than material you receive, and desirable difficulties (1992), practice that feels harder is remembered longer
sometimes a summary is all you need. What it misses when you need more, even with every sentence right, is the delta, what this work adds that the work before it did not have. A difference exists only against something, so the delta lives between documents rather than inside any one of them, in relations a single-document summary cannot carry.
the mechanism under the missing delta: the R in RAG is the vector space model (1975) with learned coordinates, retrieval ranks by similarity and generation continues toward the likely, so an unprompted summary drifts to what is typical of its sources
and similarity is always relative to an anchor. term specificity (1972): a term is specific only against a chosen collection. A reader who has read deeply carries a concept map, so who to compare against already encodes the prior; a cold question gets generic neighbors, many of them and none ranked by what your task needs
that map is also where B’s framing taste came from: both stories started after someone had framed the question, and the frame is a place on a map only readers have
so the delta is what you read for, via a four-question recipe (motivation, solution, old idea, delta); the recipe has its own ancestor, the Heilmeier catechism (mid-1970s), which asked these questions of funding proposals; we ask the same questions of technical papers, and the answer is allowed to be “nothing, honestly”

D. reading practice that works

before retrieval and before generation, the dream ‘research’ machine was one that kept a reader’s trails: the memex (1945) would store what a reader connects, so a path once found could be followed again later
one popular idea, a model-maintained wiki of your reading (the LLM-wiki pattern, 2026): each new source folded into existing pages, contradictions flagged, your own good answers filed back. The claim is compile rather than retrieve, integrate once instead of searching the pile per question; and the page is an index, its one-line entries telling you which article to open
the wiki idea is similar to the Zettelkasten, which grew to about 90,000 cross-indexed cards from 1952 onward and stood behind roughly 50 books and 550 articles, and its owner did the filing by hand for four decades. The memex was never built for the same reason nobody else copied the practice. What changed is the price of the bookkeeping
Good class notes were never a transcript; the rewriting into your own structure is what made them worth returning to, similarly for cheatsheets; the wiki is that move applied to reading
the same move at the single-paper grain: say your understanding to the model and let it correct you, then write paraphrase notes of the paper’s motivation and methods once the conversation has tested them; this is part C’s generation effect with the checking added, and the notes reread later without rereading the paper; an example from your peer (2026) alongside the conversations that produced it (architecture, guidance)
why the map matters: the same summary that leaves a novice with nothing is the mature reader’s “efficiency silver bullet”. Mature readers bring the map, so they know what to fetch. Skim first, delegate second; ask when stuck, not when starting, naming the step you cannot follow rather than the whole paper; and the figures are what a summary-only reading never sees
the memory layers now shipping in AI tools reach for the same thing; the difference is ownership. Here the reader keeps the structure, and the model gets hand-picked neighbors instead of generic ones

Page updated August 29, 2026.