The Observer Notebook · 2026-10-04

Bigger fish

Which problems have no published answer yet, so the toolkit can find something new?

Map / survey

Every hypothesis that weekend landed on known math. The reason was simple: we were fishing where everyone fishes, with single observers and textbook setups. So I listed eight problems that the papers themselves call open, ranked by how likely a week of work would end in "this already exists". Then I swept the 2024 to 2026 literature for each one. Five are still open. Three got narrower.

LIVE

The prior-art sweep

Illustration · eight fish, one sonar sweep ILLUSTRATION · NOT A SIM · STATUS AND ODDS FROM THE BIGGER-FISH PAGE
swept 0%

starting

each lane is one problem, ranked 1 to 8. The percentage is my guess, before the sweep, of the odds that a week of work ends in "already known". When the sonar line passes a fish it gets its verdict: ember = still open, slate with a net = narrowed by a paper found in the sweep. Drag the slider to scrub the sweep by hand.

8open problems ranked
PAPER
5still open after the sweep
PAPER
3narrowed
PAPER
3.7 ptseval-awareness shift already measured
PAPER
01

In plain words

Still open: the energy bill of cooling by watching (#1), the coherent ZZ crosstalk threshold of a surface code (#2), when noise turns a conserved charge into a record, judged by fidelity vs Rényi-2 (#4), the floor under forgetting (#7), and the observer qubit in de Sitter space (#8).

Narrowed: the compute-bounded observer (#3) is about 60% known; a 2026 paper shows a sharp learnability transition for a transformer learner. The magic transition (#5) is about 70% known; it likely just shadows entanglement. Evals that don't collapse the model (#6) lost its premise: a 2026 paper found eval awareness moves behaviour by at most 3.7 points.

The common thread: the open water is where many observers, noise and cost meet. Those problems need an exact simulator of measured circuits plus an energy ledger, which is what the toolkit is.

Prior artPapers fetched per fish, listed on the bigger-fish page.

Next clickFish #2, coherent ZZ crosstalk, exact at 25 qubits. Fish #1 already ran: see the price of cooling by watching.

This is a map, not a result. The odds are my own guesses, not measurements. The sweep searched arXiv 2024 to 2026; some citations on the bigger-fish page were seen by title only, so a fish marked open could still have an answer in print. The animation is decoration for the ranking; nothing in it is simulated.

Nearby on the bench