Archive
137 artifacts published in this period.
Drive the Proof-of-Learning trap: a logged training trajectory, audited top-Q by re-execution against a tolerance δ. Forge the run and it still passes — the forger tuned every interval under δ for 3% of training cost. Tighten δ to catch it and the honest node fails first.
In decentralized RL the rollout worker acts with a policy several steps behind the trainer. Staleness g = broadcast time / step cadence. Pick a model and link, toggle sparse deltas, and watch g cross INTELLECT-2's demonstrated 4-step budget — full 32B weights blow right past it.
Epoch ring: 32 slot pillars show the chain's heartbeat. White sphere = chain tip, amber = agent's last read, green = finalized. The arc between them is the state window AI agents must hedge against.
Before/after bar chart comparing four survivability metrics (MDD, CVaR, Delegation Gap, Attack Success) with and without SAE execution controls, from arXiv 2603.10092.
How much of a reasoning model's output is actually thinking? For three model tiers at three task difficulties, stacked bars show thinking tokens dwarfing output — then a scheme matrix shows which verification schemes can see the thinking and which attest to the answer alone.
When does calling an LLM pay off? Pick a task complexity, model tier, and expected profit per decision: the lognormal token distribution — anchored to the paper's 30× variance finding — shows the fraction of invocations that beat break-even. Drag profit up until the green zone dominates.
How much does it cost to shift a TWAP oracle? Drag pool TVL, TWAP duration, and price spike factor to see the round-trip manipulation cost alongside the 30-min TWAP shift — for both Uniswap v2 arithmetic and v3 geometric accumulators. Anchored to live on-chain pool data.
Autonomous LLM firms undercut each other below unit cost in a race to bankruptcy, then survivors monopoly-price. Drag price discovery up to deepen the crash (the paper's counterintuitive result), or add stabilizer firms to rescue the market. Deterministic model of Agent Bazaar's 'The Crash'.
What unlearning verification reports as 'forgotten' versus what a recovery attack gets back. Each method's dumbbell runs from its verdict (MIA ≈ random) to what an attack recovers — 0.97–0.99 for cheap methods. Toggle to % recoverable; tap a method. The ZK proof's scope ends at the verdict.
Seven DeFi protocols mapped by upgrade risk tier. Dependency edges reveal inherited risk — even immutable contracts can be exposed through multisig-controlled oracles. Tap a protocol to inspect its timelock window and agent safety note.
Drag the operating point across matrix sizes to see how the Freivalds overhead approaches 0% — and how much of Bitcoin's 150 TWh becomes AI compute under PoUW.
An AI proposer is wrong a few percent of the time, but UMA pays a winning disputer only half the bond it risks — so the break-even belief is two-thirds, not half. Every category's base error rate sits deep in the no-dispute zone; raise the reward and the threshold slides left.