The goal is creating autonomous agents, eliminating human intervention

2024-10-22 14:07:12Lifestyle SHKRUAR NGA REDAKSIA VOX

The goal is creating autonomous agents, eliminating human intervention

Facebook owner Meta has announced that it is introducing a series of new artificial intelligence models from its research division, including an "evaluator" that could offer a path to less human involvement in the development of innovative technology.

This technique involves breaking down complex problems into smaller steps and improves the accuracy of answers to challenging problems in subjects such as science, coding and math.

It is reported that the Meta researchers used entirely AI-generated data to train the estimator model, also eliminating human input at that stage.

The ability to use the new technology to reliably evaluate intelligent models paves the way to building autonomous artificial intelligence agents that can learn from their mistakes, the two Meta researchers said. As reported, this sector envisages agents such as digital assistants intelligent enough to perform a variety of tasks without human intervention.

Self-improving models can cut the need for an expensive and inefficient process used today, called Reinforcement Learning from Human Information, which requires input from human markers who must have specialized expertise in the field.

"We hope, as the technology becomes more and more super-human, that it will get better at controlling its own work so that it is more capable than the average human," said one of the researchers.

Other companies, including Google and Anthropic, have also published research on the concept of reinforcement learning by specialized humans. However, unlike Meta, these companies tend not to release their models for public use.


Video