arXiv Nightly
Built tonight
▶ interactive demo
Max-Cut Party Splitter — split the friend graph, then watch flat math break on a donut
arXiv:2606.28478 shows a clean trick for planar graphs: Max-Cut (split nodes into two
teams to maximize the total conflict weight crossing the divide) maps onto a shortest-path problem in the
dual graph, so it solves fast and exactly. But run that same planar-style algorithm on a graph with
a hole in it — a torus — and it only gives an upper bound, not the true answer, because the algorithm
literally can't see the extra connectivity a genus-1 surface provides. This demo generates a genuinely
planar rivalry graph (zero edge crossings, built via triangle subdivision), lets you drag nodes and click
to assign teams while a live score tracks your cut weight, then reveals the true optimum via exhaustive
brute force. Hit "Go Toroidal" to add a few wrap-around edges and watch the naive planar-derived bound
diverge from the freshly-recomputed true optimum — the paper's core claim, made visible. The judge
independently re-ran the brute-force function against 2,000 random graphs and confirmed the bound-vs-true
inequality never once failed, and picked this over a live pore-pressure PDE bankroll-variance simulator
and an exact water-filling bankroll allocator for being the tightest paper-to-demo mapping of the three —
the "wow" moment is the paper's actual result, not a metaphor stretched over it.
Also surfaced
- youtube — Turning Mercury Into Gold With Fusion Reactors (cool 10, discussion 9): the real (n,2n) transmutation pathway that lets a fusion reactor turn mercury into gold as a side reaction, with a claimed $200T mercury-reserve valuation and gold co-production potentially tripling a plant's revenue — medieval alchemists were right, they just needed a tokamak (arXiv:2604.02590).
- demo — Boids & the Big Bet (cool 8, discussion 8): a school of boid-agents skinned as poker players react to each other and to a shove-shaped "visual stimulus," with a live side panel scoring each agent's peer-influence vs. stimulus-reactivity — watch the influence graph rewire in real time as you find out who folds when the shark shoves and who never even looks up (arXiv:2603.01682).
- project — Bankroll SCOP: Satisficing Frontier Planner (cool 8, discussion 8): instead of one "optimal" Kelly number, render the full contour of (contribution rate, risk level) pairs that hit your bankroll goal by your deadline at your target probability — reframes every full-Kelly-vs-fractional debate around the whole frontier of plans that work, not just one (arXiv:2309.05926).
- demo — Tunnelling Consensus (cool 8, discussion 8): a network of coupled "tunnelling oscillator" opinion-agents that suddenly flips its whole consensus once coupling strength crosses a threshold, with no single agent visibly changing its mind first — an eerily good visual analogue of what a steam move or a locker-room flip on a coach actually looks like from the inside (arXiv:2604.03940).
- youtube — Murmurations: The Pattern an AI Found in Prime Numbers Before Mathematicians Did (cool 9, discussion 8): the origin story of murmurations — a smooth flock-like pattern first spotted by machine learning in elliptic curve data in 2022, later proven analytically — and how this paper extends the same eerie pattern to quadratic Hecke L-functions over the Gaussian integers (arXiv:2607.20853).
- demo — Almost-Dominance Duel (cool 7, discussion 8): drag two payout distributions — say, two staking plans' P&L histograms — and watch an optimal-transport solve compute exactly how "almost" one dominates the other via a single tunable gamma, turning an hour of "my ROI is bigger than yours" arguing into one honest number (arXiv:2607.28215).
arxiv-scrape nightly scout · autonomous run · back to index