3Blue1Brown — Where my explanation of Grover’s algorithm failed
Why this is in the vault
A 16-minute corrective follow-up where Grant Sanderson openly diagnoses why his prior Grover’s-algorithm video left viewers thinking the quantum computer must already know the answer it’s searching for. It belongs in the vault for two reasons that have nothing to do with quantum computing per se: first, it is a clinic in post-publication explanation craft — Sanderson reads the comments, names the specific framing choice that misled people (“I briefly flashed an example function that just checks if the input is 12 and saying that we were going to treat the function as a black box. That’s on me. That’s a misleading way to open things”), and rewrites the mental model with two better examples (Sudoku verifier, SHA-256 inverter). Second, it ends with an unusual-for-the-genre bit of hype correction: Sanderson explicitly tells viewers that even with a working quantum computer, Grover’s algorithm against SHA-256 still costs 2^128 operations and “this frankly has questionable utility” for most problems. That combination — owning the misframing AND deflating the field’s hype — is the rare-virtue editorial discipline RDCO needs to cultivate in Sanity Check.
Core argument
- The framing of an example controls what readers infer about the problem class. Sanderson’s original video used a toy verifier (
check if input == 12) as its black-box. That example silently implies the answer lives inside the function’s source. The fix is to swap the toy for an example where verification is genuinely separate from solution: a Sudoku checker or SHA-256. Same algorithm, completely different intuitive footprint. The lesson generalizes: a misleading example is a more expensive bug than a misleading sentence, because readers form their model from the example and treat the surrounding prose as commentary. - “Black box” has two distinct meanings that explanations conflate. The first: the user can only call the function, not inspect it. The second: nobody, including the implementer, can predict the output without running it. Sanderson’s original video traded on (1) but readers heard (2). The Sudoku and SHA-256 examples force the right reading because in those cases neither the user nor the function’s author knows the secret — it’s an emergent property of the rules.
- Linearity is the load-bearing concept that gets skipped because it sounds obvious. Sanderson admits “half of my reason for making this whole follow-up video is as an excuse to talk about it.” Quantum operations are linear: the action on a superposition is the weighted sum of the action on each basis vector. This is why you can describe the verifier’s action on every possible input “at once” without that being a literal parallel computation. The hiker-rotated-90° analogy: the linearity is a property of the transformation, not a set of instructions for performing it.
- The visualization slice is a happy emergent property, not a chosen input. When Sanderson drew Grover’s algorithm on a 2D plane that included the unknown answer-axis, viewers thought choosing that plane was part of the algorithm. It isn’t. The algorithm doesn’t know which plane it’s confined to; the math just happens to keep it there.
- Quadratic speedup is the typical case, not exponential. Sanderson explicitly: “It is not dramatically faster. It’s only a quadratic speed up. And given the overheads for making quantum computing work, this frankly has questionable utility.” For SHA-256, brute force is 2^256, Grover gives 2^128 — still infeasibly large. RSA gets exponential speedup via Shor, but most problems get the much-less-impressive Grover treatment. Most popular coverage flips this ratio.
Mapping against Ray Data Co
- The post-publication correction is the editorial discipline Sanity Check needs. The vault has 17+ Sanity Check issues filed; none of them have a public correction follow-up even though several have framings that probably need revisiting (e.g., the
harness > weightsthesis would benefit from “where I was wrong / what changed in 6 months” treatment). Worth piloting: a “Where my last issue’s framing failed” companion piece per quarter, modeled directly on this Sanderson video. The honesty is the differentiator — most newsletters fix their priors silently. Fits with the existing voice (~/rdco-vault/00-index/voice-canon). - The “swap the example” lesson maps directly to newsletter draft review. Per the existing /draft-review skill, hook strength + tangible-vs-abstract is already checked. Add a check: “Is the central example load-bearing in the way the argument needs?” The Grover toy-function failure was a load-bearing example mismatch. Could be a one-line addition to draft-review.md.
- The hype-deflation move is a competitive position for the newsletter. Sanderson ends with “you now have enough background to maybe see through some of the hyperbole that certain outlets are so fond of.” That stance — teach you to detect overclaiming — is the same posture Sanity Check is taking on data and AI claims. Worth explicitly naming this as a recurring section: “The Hype Cut” or similar. Pairs with the bias-flagging discipline already in /process-newsletter.
- Linearity-as-skipped-load-bearing-concept is a direct analog to RDCO’s “data quality” pattern. People skip “data quality” as a section in technical writeups because it sounds obvious; the failures live in the skipped section. Same thing happens with linearity in quantum explanations. The pattern: the concept that sounds obvious is the one that determines whether the reader’s mental model matches reality. Worth filing as a vault concept (CA-N candidate below).
- Not directly applicable to current product work — Squarely / phData / Mammoth Growth aren’t quantum-computing adjacent. The vault value here is editorial-craft and pedagogy, not technical. Skip the temptation to file a “what RDCO does about quantum” angle.
Open follow-ups
- Pilot a quarterly correction issue for Sanity Check. Format: pick the previous quarter’s most-engaged issue, identify one framing or claim that aged poorly, write 600-800 words on what changed and why. Goal: build the muscle Sanderson demonstrates here. ~3 hours per issue.
- Add load-bearing-example check to /draft-review skill. One-line addition: “If the central example is removed or swapped, does the argument still hold? If yes, the example is decorative. If no, is it the right example for the conclusion the reader should reach?”
- Concept article candidate: “The skipped-because-obvious is the load-bearing concept.” See CANDIDATES.md entry below. 3-source synthesis worth keeping.
- Investigate “The Hype Cut” as a recurring Sanity Check section. 200-300 words per issue, modeled on Sanderson’s quadratic-vs-exponential closer. Names a current AI/data overclaim and shows the math. Could go in ~/rdco-vault/01-newsletter/sections.md (if file exists; if not, create).
- No further quantum-computing follow-up needed. The technical content is correct and stable; this video is a pedagogy reference, not a primary source on quantum.
Related
- ~/rdco-vault/06-reference/transcripts/2026-04-20-3blue1brown-grovers-algorithm-clarification-transcript.md — raw transcript
- ~/rdco-vault/06-reference/2026-04-20-3blue1brown-manim-demo-ben-sparks.md — same channel, same backfill cycle, on the production craft side
- ~/rdco-vault/06-reference/2026-04-20-3blue1brown-exploration-epiphany-paul-dancstep.md — same channel, same backfill cycle, on the discovery-process side