Summer 2021 witnessed the quiet closure of OpenAI’s mulrobotics team due to insufficient data required for training robots in AI-based motion and reasoning. However, this challenge seemed to be overcome by Covariant, a startup established by three of OpenAI’s early research scientists in 2017. They have introduced a system that integrates the reasoning capabilities of large language models with the mechanical dexterity of an advanced robot.
Named RFM-1, the model was developed leveraging years of data amassed from Covariant’s array of item-picking robots, supplemented by words and videos sourced from online platforms. The system can be prompted using five distinctive inputs: text, images, video, robot instructions, and measurements. Covariant anticipates the system’s proficiency and efficiency to enhance as it is being applied in the real world.
On another note, last week saw the implementation of a long-awaited set of climate rules by the US Securities and Exchange Commission. The ruling mandates the majority of publicly traded corporations to disclose their greenhouse-gas emissions in addition to the climate risks accumulating on their balance sheets. Regrettably, the federal entity diluted the ruling due to heavy lobbying from commercial sectors, thereby restricting its overall effectiveness. This was considered a missed opportunity for compelling companies to confront the escalating perils of global warming.
