OpenAI reportedly now has a five-level scale that determines its progress in building artificial intelligence models capable of outperforming humans. The company claims it is now near level two.
According to an OpenAI representative, the inventor of ChatGPT, who is recognized as a leader in the race to create ever-more-powerful AI systems, introduced the new classification method to the team on Tuesday during an all-hands meeting.
(Photo : Photo by KIRILL KUDRYAVTSEV/AFP via Getty Images)
OpenAI plans to reveal the levels to possible investors and outside stakeholders. They range from Level 1, which depicts the present state of AI technology that can communicate verbally with humans, to Level 5, which represents AI that can perform organizational duties.
The spokesperson claimed that OpenAI executives told employees that the company is nearing the second level, or "Reasoners"-systems that can solve regular problems in addition to a doctorate-level trained person without the need for tools that they are currently at the first level.
Read Also: Apple, Microsoft Abandon OpenAI Board Seats Amid AI Regulatory Scrutiny
Latest GPT-4 Developments
At the same meeting, business leadership presented a research project involving its GPT-4 AI model. According to an anonymous source familiar with the proceedings, OpenAI claims to display some new skills leading to human-like reasoning.
The source was not authorized to speak to the media. When contacted for comment regarding the demonstration, the representative stated that OpenAI regularly tests new capabilities internally, which is standard procedure in the sector.
For many years, OpenAI has stated that it is developing artificial general intelligence (AGI), or computers capable of performing most activities more efficiently than humans. These kinds of systems don't exist yet. As previously stated, Sam Altman of OpenAI believes that AGI may be achieved this decade.
Nvidia's Jensen Huang on AGI
OpenAI's ranking system could be one of the first objective ranking systems for a company's progress towards AGI, as even Nvidia's Jensen Huang claimed that achieving AGI depends largely on what it means.
At the beginning of March, Jensen Huang claimed that AI might someday think like humans, although he clarified that this would depend on exactly what he meant by that. He asserts that AGI might occur in as little as five years if it meant passing most human testing.
According to Huang, the leader of the world's leading producer of AI chips, which power systems like OpenAI's ChatGPT, AGI might materialize in five years if AI is given every test imaginable and does well on each one.
But depending on other standards, he contended, AGI might be much closer given the ongoing disagreement among scientists over the most accurate method to describe human thought processes.
According to reports, demand for AI chips is expected to rise further, making AGI feasible. However, Huang asserts that the computational capacity of AI chips will rise in tandem with demand.
Related Article: Grok AI Delivers Near-Instantaneous Responses, Outpacing Common Models
