I've built an evaluation framework measuring whether frontier AI models respond fairly across different value perspectives. It’s inspired by Community Notes after reading an article by @audreyt, see below. She argues AI alignment can't be top-down. We need "uncommon ground", responses people with different views can all find reasonable.
So far the framework uses AI personas as raters, and they're showing all frontier models tested lean progressive. I'm now validating this with real humans, to test whether the AI personas can reliably represent diverse human values when evaluating AI models. If you've got 5 mins, please help me out. Rate a few AI responses and see how your values compare: https://t.co/nCczW9W0ZP
Audrey’s article: https://t.co/6h64LhqA7a
Research: https://t.co/hhXQx5T4q4
@zksync, @the_matter_labs’s zkRollup for trustless, scalable payments is now live on Ethereum mainnet!
Give it a try: 👉👉 https://t.co/plxFMLdukr
Claiming my trial tokens: 3507692160956296
@AaronFosterReal @OKDunc @UniswapExchange In this dir: https://t.co/psEFN7PjK0 I reckon you could just do:
$ truffle migrate --network kovan
$ truffle exec scripts/addTokenExchange.js --network kovan
See readme: https://t.co/gzsPBUuN8f