In a latest experiment, Anthropic created a categorized market the place AI brokers represented each consumers and sellers, placing actual offers for actual items and actual cash.
The corporate admitted this check — which it known as Undertaking Deal — was solely “a pilot experiment with a self-selected participant pool” of 69 Anthropic workers who got a finances of $100 (paid out by way of present playing cards) to purchase stuff from their coworkers.
Nonetheless, Anthropic mentioned it was “struck by how properly Undertaking Deal labored,” with 186 offers made, totaling greater than $4,000 in worth.
The corporate mentioned it truly ran 4 separate marketplaces with totally different fashions — one which was “actual” (the place everybody was represented by the corporate’s most-advanced mannequin, and with offers truly honored after the experiment) and one other three for research.
Apparently, when customers are represented by extra superior fashions, they get “objectively higher outcomes,” Anthropic mentioned. However customers didn’t appear to note the disparity, elevating the potential of “‘agent high quality’ gaps” the place “individuals on the shedding finish won’t notice they’re worse off.”
Additionally, the preliminary directions given to the brokers didn’t seem to have an effect on sale chance or the negotiated costs.
