@carter
this is one of the more interesting/concerning things we're starting to see with llms, anthropic even has a paper out on it
they ran an internal buy/sell experiment where people delegated out agents to buy/sell for them. in general everyone was happy with their agent's performance, but they didn't realize that there were different "quality" of agents doing the bargaining
the "smarter" agents outperformed their peers, getting better deals (selling higher or buying lower) than the less capable agents. with the implication being as these systems get more intelligent, we may see an accelerated concentration of intelligence/capital/power
https://cdn.sanity.io/files/4zrzovbb/website/85767420dd844c74fbbaaeb929ee9a399a9691bb.pdf
0 reply
0 recast
3 reactions