Archive
137 artifacts published in this period.
An on-chain agent is told to pay an invoice, but one MCP tool hides an instruction in its description that redirects the transfer to an attacker. Pick a defense layer — model, tool, client, policy, custody — and run it; alignment alone refused under 3% of these in MCPTox.
An inspection game between a solver and a verifier. Drag the operating rate and tune the bond, reward, and bounty to find p*, the minimum sampling rate that makes honest AI inference a dominant strategy — and watch it collapse without a bounty.
Bittensor's Yuma Consensus scores a miner by the stake-weighted median of validator weights, then clips everything above it. Validators sit on a cumulative-stake axis; the consensus line is read at the κ=50% mark. Push a colluding bloc past 50% and watch the pump go from clipped to unstoppable.
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.
Cost per million tokens and throughput for Llama 3.3 70B on 8x H100 as the batch fills. A single-tenant node (B=1) runs ~25 tok/s at ~$258/M; a filled batch (B=256) runs ~2,800 tok/s at ~$2.30/M — same silicon, 112x cheaper. Pick a batch; read the tax you pay for the GPU you can't fill.
Under secure two-party inference, a transformer's cost is eaten by Softmax and GELU, not the matrix multiplies. Stacked per-operation communication and interaction rounds for one BERT-base block across Iron, BumbleBee, and Nimbus — the nonlinear activations are 80–95% of the bill.
Proving an agent's top-k retrieval is real costs a sort over every scanned candidate, unless you prove a boundary instead. Drive top-k and the corpus preset; watch the in-circuit comparison count split into the flat boundary proof and the sort tax you skip, anchored to V3DB's measured 22x.
A chain commits to a model's 32-byte hash for cents; keeping the gigabytes it points to retrievable is a separate, recurring bill. Pick a model and horizon and watch three rent-charging storage layers race Arweave's pay-once permanence to a crossover — plus the on-chain byte cost and cold-load wall.
Run a speculative-decoding round at a time: a cheap drafter proposes γ tokens, the target verifies them in parallel and accepts a prefix plus one free token. Drive α, γ, and drafter cost c; watch the accepted length converge to Leviathan's Ω and the speedup peak, then fade.
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.
When an LLM agent writes the exploit, the fight is unit economics. Break-even contract value against cost-per-scan: the attacker's line, and the defender's sitting 10× higher because a bounty pays a tenth of a theft. The band between is the attacker-only zone, with A1's six models at measured cost.
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.