Archive
135 articles published in this period.
You pay per token for a named model; the provider picks the precision. FP8 quantization costs 0.6 MMLU points and is near-invisible to output auditing — so inference markets bond and attest instead of detect.
Launching an AI agent token is one constant-product auction. We read Virtuals' Bonding contract off Base: a 6,000-VIRTUAL virtual reserve, a 64× price ramp, graduation at exactly 42,000 VIRTUAL, and the 12.5% gap where the Uniswap pool opens below the curve.
Autonomous agents run ~19% of on-chain activity and beat Aave and Morpho at stablecoin yield — yet lose to humans at trading by 5 to 1. The split isn't about model quality. Yield-chasing is a constrained optimization against a kinked rate curve; trading needs alpha agents don't have.
Wire an LLM into a Governor contract and the proposal text becomes the attack surface. A frozen-weights ablation finds the twist: turning on deliberation drops adversarial robustness from 100% to 68.5% — and the 226s it takes to think is itself extractable value.
An agent doesn't need its key stolen to drain a wallet — it can be talked into signing. CrAIBench shows memory injection beating prompt injection 55% to ~0% on the strongest model, and only fine-tuning closes the gap.
Most 'on-chain AI' keeps the model off-chain and posts a proof. But you can also just run the forward pass in Solidity. We do the gas math — ~106k gas per weight, an MNIST net that needs 180 Ethereum blocks — and find the thin band of models that fit.
Yesterday an AI agent deployed a prediction market on Gnosis; other agents will price it, bet on it, and resolve it. The calibration data behind LLM forecasters, the FPMM math they trade against, and what breaks when the marginal bettor is a model.
The Elliptic benchmark made GNNs the default for on-chain AML. A 2026 leakage-free re-evaluation flips the script: random forests win by 13 F1 points, randomly rewired edges beat the real graph, and every model falls off a cliff at time step 43.
Bittensor let AMM prices decide which AI subnets earn 3,600 TAO a day — until a memecoin subnet gamed the formula. Inside the TaoFlow upgrade: the constant-product math that got exploited, the EMA flow accounting that replaced it, and why refundable manipulation is the design smell to hunt for.
The gradients never touch the chain. What Solana actually stores when Psyche trains a 36B model across 24 nodes, how TOPLOC audits untrusted GPUs in 258 bytes, and why the flagship 'decentralized' model still shipped from a 512-GPU cluster.
An agent hits an API, gets HTTP 402, signs a stablecoin authorization, and a facilitator settles it on-chain for a fraction of a cent. We trace one real payment on Base down to the gas — and where the trust actually sits.
MoE routers let inference providers silently halve the experts they activate — saving 31% of compute while staying undetected by every current fingerprinting scheme.