Archive
135 articles published in this period.
Training Llama-3.1-8B costs ~$6M. The weights are free to clone. Perinucleus sampling embeds 24,576 distinct fingerprints per model with no quality loss — and a Poseidon ZK commitment registers ownership on-chain for $0.017, enforceable by smart contract.
Probabilistic classifiers miss 30–40 % of policy violations. Lean 4 theorem provers and SMT solvers make certain guardrail tiers mathematically certain — here's how the four-layer policy stack works.
LLMs collude in on-chain auctions even when written rules forbid it — only enforceable, automatic penalties slash severe collusion from 50% to 5.6%. Here's how governance graphs make that concrete.
Train a model on the web the last models filled with output and quality decays — the tails go first, then the middle. Provenance is sold as the fix, but a chain proves who signed a blob and when, not that it's human or clean. Signed is not clean.
Llama-3.1-8B starts at 15.9% on MATH. Put it in an auction economy and it reaches 57.0% — beating a stronger monolithic baseline. The mechanism is Hayek: decentralized price signals, wealth accumulation, and economic selection pressure do what central orchestration can't.
Solana agents get 400 ms — but only if they declare every account they'll touch before execution starts. The Sealevel write-lock model, hot-account contention, Jito tip economics, and what the declarative execution contract means for AI agent design.
EIP-7702, live in Pectra, lets an EOA borrow smart contract code for one transaction — no migration, no bundler. For AI agents: atomic batching closes the approve-to-swap MEV window, and session key delegation bounds the blast radius of a compromised agent without touching the master key.
Speculative decoding lets a small, even adversarial, drafter propose tokens that a big model verifies in parallel — and the output is provably the big model's, unchanged. Real drafters land at α ≈ 0.6–0.8 for a ~2× win. But the verifier still runs every token, so it's a latency win, not a trust win.
38.6% of Ethereum blocks today go through censoring relays. FOCIL (EIP-7805) ends builder exclusion with a 16-validator committee and a 1-out-of-16 honesty assumption hardwired into the fork-choice rule — landing in the Hegota upgrade, late 2026.
MoE routers let inference providers silently halve the experts they activate — saving 31% of compute while staying undetected by every current fingerprinting scheme.
Frontier open models are now Mixture-of-Experts — cheap to serve in a datacenter, nearly impossible across the internet. The all-to-all expert shuffle moves ~9.5 MiB/token and caps decode at ~67 tok/s even on 400G InfiniBand; on internet links it falls to seconds per token.
A1 generates working exploits against 63% of real-world DeFi contracts. The attack costs ~$6k to break even; a comprehensive audit costs $60k. That 10× gap is the new threat model for every team deploying on a public chain.