> Some companies believe they can scale the data wall by using synthetic data — that is, data that is itself generated by A.I. systems — to train their models. But many researchers doubt that today’s A.I. systems are capable of generating enough high-quality synthetic data to replace the human-created data they’re losing.
I'm thinking synthetic datasets are how it's going to go. Of course you can't get information from nothing, but they might not need nearly as much seed data to generate lots of examples that are specifically designed to train reasoning skills.
Maybe it will take a while to get over the hump, but they'll be motivated to make it work.
I'm thinking synthetic datasets are how it's going to go. Of course you can't get information from nothing, but they might not need nearly as much seed data to generate lots of examples that are specifically designed to train reasoning skills.
Maybe it will take a while to get over the hump, but they'll be motivated to make it work.