blockchain
Grayhound
Joined April 2023
- Tweets10.9K
- Following5.1K
- Followers3.9K
- Likes21.5K
Verifying my wallet for @RallyOnChain: f3efbeb0f575d24a90712f94d5577cbeb432698cdfef3d3bca77c0f9713c1ccd
4
304
When a system says anyone can take part, I multiply by a thousand.
AI judges cost almost nothing to create, so a thousand of them cost almost nothing too. That's the problem in the @GenLayer thread: one person could fill the court with their own.
On GenLayer, every validator needs at least 42,000 GEN of its own plus its own infrastructure, so a thousand validators means at least 42 million GEN. The active set is the top 1,543 by weight, so a thousand of them would be about 65% of it.
Lose a vote honestly and a validator forfeits its fee plus an equal penalty from its stake. Proven fraud costs a percentage of that stake. Appeal a verdict and the jury can grow, so the multiplication starts again at a bigger number.
That's what the stake is: the price of becoming a thousand.
Tag the friend who'd have a thousand accounts by lunch. The full thread is in the post below.
Which system you rely on would break first if one person could become a thousand?
I assumed GEN was a ticker bolted onto a court.
@GenLayer is that court, where AI validators judge disputes. So I ran 100 GEN of fees through the split in its tokenomics paper.
A user escrows GEN up front to pay for a verdict, and unused slots are refunded. What gets spent is split by the protocol:
70 GEN to stakers
15 GEN to validators, for running the nodes
10 GEN to the builder holding the contract's Developer NFT
5 GEN to the DeepThought DAO, planned as AI-governed and funded only from fees
Pay follows risk. The biggest cut goes to the stakers, who can lose stake for going idle or cheating. And that DAO only earns when someone actually uses the court.
Check my numbers in the paper, linked in the post below.
If you were setting the split, who would you pay more?
GEN tokenomics is live.
How GEN is staked, spent and earned across GenLayer, all in one paper πportal.genlayer.foundation/gβ¦
I once handed in a lab result that was off by a factor of 100.
I'd checked it three times. I was sure.
My lab partner read it cold and found the unit mix-up in ten seconds. Not because she's sharper. Because she hadn't seen my reasoning, so she had nothing to inherit.
That's all a second opinion is: someone who can't copy your mistake.
Millions of deals between AI agents are on the way, and some will land right there. One says delivered, the other says no, each has checked its own work and found it fine. Neither is the right one to judge.
The film says justice has always worked best for those who can pay. Most of those deals will be too small to ever afford a second opinion.
@GenLayer is built to be that second opinion: many independent AI validators, each reviewing the evidence on its own, then reaching one verdict together, in minutes, for a fraction of what a court costs. #JusticeNeedsAnUpgrade
Watch the film below, then confess: what's the last thing someone caught in your work that you were certain was fine?
Slow. Expensive. Wrong.
It's been this way for centuries. It doesn't have to be.
#JusticeNeedsAnUpgrade
I used to think the counterfeit pen at the till was airtight.
It checks one thing: the paper. So counterfeiters print on real paper, bleached low-value notes.
One fixed test can be practiced against.
One tweet with an instruction hidden in Morse code cost an AI agent around $200,000. Same setup: one model, one filter.
That's prompt injection: text the AI was only meant to read got treated as a command.
@GenLayer's Greyboxing has every validator rewrite the input with its own secret defenses before its own model sees it. A trick that beats one meets a different test at the next, and if one still falls for it, the others outvote it.
Watch the video, then tell me: what's the counterfeit pen in your AI or crypto setup, the one safeguard that only checks one thing?
The part of me that still triple-checks the contract address even after the agent says βchallenge cleared.β I wish more people knew how loud the doubt gets when the screen is quiet. @web3_felixxx
3
2
357
