Kleros study pits AI models as jurors on real disputes
Kleros researchers Federico Ast, William George, and Rob Dean re-ran 99 fintech disputes that Kleros had already resolved, with an LLM acting as juror. ChatGPT ruled for the consumer 3 times in 99 cases, while Claude did so 14 times. The study was published on the Kluwer Arbitration Blog (Wolters Kluwer).
Sources
- Federico Ast, William George and Rob Dean re-ran fintech disputes Kleros had already resolved, with an LLM in the juror's seat. ChatGPT found for the consumer 3 times in 99, Claude 14. On the Wolters Kluwer arbitration blog. https://t.co/BA0qyWEi1t Quoting @Kleros_io: 🧮 Which model would you want to judge you? Kleros researchers just put numbers to that question in a study published on the Kluwer Arbitration Blog, one of the most-read venues in international arbitration. They took 99 real fintech disputes, all previously decided by human https://t.co/7VJfuqL9iM Replying to their earlier post: All three parts of How a Kleros Enterprise Case Works are out, following one dispute from intake and anonymization through to the ruling and the appeal either side can still file. https://t.co/L3yRzR…x.com ↗
- The full study, by @federicoast, @williamhwgeorge, and Rob Dean, on the Kluwer Arbitration Blog (Wolters Kluwer): https://t.co/S5bqVAo90t Replying to their earlier post: 🧮 Which model would you want to judge you? Kleros researchers just put numbers to that question in a study published on the Kluwer Arbitration Blog, one of the most-read venues in international arbi…x.com ↗
- 🧮 Which model would you want to judge you? Kleros researchers just put numbers to that question in a study published on the Kluwer Arbitration Blog, one of the most-read venues in international arbitration. They took 99 real fintech disputes, all previously decided by human https://t.co/7VJfuqL9iMx.com ↗
Kleros
133 followers
A decentralized arbitration protocol for disputes in the #web3 economy. Settled 1200+ disputes and counting! We're Hiring! http://kleros.io/career
