Blokz Development Co.
Deep technical writing and interactive artifacts on AI × blockchain — from neural nets to consensus, agents to smart contracts, and the infrastructure binding them together.
Journal
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.
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.
Majority voting over LLMs throws away the one node that got it right. Fortytwo's swarm inference ranks answers pairwise instead — +17 points on GPQA Diamond — with on-chain reputation and proof-of-capability for Sybil defense. The mechanism, the math, the tradeoffs.
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.
Every on-chain scheme that verifies an LLM by re-executing it and comparing digests assumes a forward pass is bit-for-bit reproducible. It isn't — Thinking Machines got 80 different answers to one prompt at temperature 0. Here's why, and what determinism costs.
DX Terminal Pro ran 3,505 LLM agents trading real ETH on Base for 21 days: 7.5M invocations, ~$20M volume, 99.9% settlement. The reliability came from the operating layer around the model, not the weights — here are the numbers and the failure modes.
Interactive
Walk the real bonding curve that prices a tokenized AI agent: 6,000 virtual VIRTUAL against 1B tokens on a constant product, spot ramping 64× to graduation at 42,000 VIRTUAL. The shaded area is VIRTUAL raised; toggle the halfway split to see the first 21,000 buy ~89% of supply.
One hard question, a swarm of LLM nodes, two ways to agree. Run a round: majority voting picks the most common answer while peer-ranked Bradley-Terry consensus surfaces the best one, and the scoreboard converges on the gap Fortytwo measured. The evaluation-edge slider shows why.
Why re-executing an LLM doesn't reproduce it. One matmul reduces to the logit gap between 'Queens, New York' and 'New York City'; drag the production split-K and watch IEEE-754 rounding flip the token at split 5 and 6, breaking the verifier's digest check. Batch-invariant kernels turn it green.
Every article and artifact in this blog lives in a public repo. Follow the work, fork the experiments, or get in touch.
30
Articles
31
Artifacts
16
Topics