OpenAI is positioning itself as a frontrunner in the quest for human-level artificial intelligence, reportedly being four steps away from achieving this milestone.
The company recently shared a five-level framework developed to monitor its progress towards artificial general intelligence (AGI) with employees, according to a spokesperson who spoke to Bloomberg. These levels range from the current conversational AI to AI capable of performing tasks equivalent to those of an entire organization. OpenAI plans to reveal these levels to investors and external stakeholders.
At present, OpenAI executives believe the company is at the first level but approaching the second, termed “Reasoners.” This level is characterized by AI that can handle basic problem-solving on par with a human holding a doctorate degree, albeit without access to tools. The third level, known as “Agents,” involves AI that can carry out various tasks for several days on behalf of its user. The fourth level, “Innovators,” refers to AI that can aid in creating new inventions.
OpenAI also showcased a research project utilizing GPT-4, demonstrating human-like reasoning skills, as reported by Bloomberg, citing an anonymous source. The company declined to provide further comments.
The framework was reportedly developed by OpenAI executives, who may adjust the levels based on input from employees, investors, and the company’s board.
In May, OpenAI disbanded its “Superalignment” team, which was focused on addressing AI’s existential risks. The company stated that the team’s responsibilities would be integrated into other research initiatives within OpenAI.