Microsoft and OpenAI have reached a consensus on a new definition of artificial general intelligence (AGI). This agreement is significant for their future collaboration and development.
A recent report indicates that OpenAI believes it will have achieved AGI only when it has created a system capable of generating $100 billion in profit. For Microsoft, the definition of AGI is not just about technological breakthroughs but also closely tied to commercial interests. According to a clause in their agreement, once OpenAI achieves AGI, Microsoft will no longer have access to the startup's most powerful models. This clause introduces uncertainty into Microsoft's future plans.
However, there are reports that OpenAI is trying to remove this restrictive clause from the agreement. According to the new AGI definition, OpenAI still has a long way to go to achieve this goal. In the current year, OpenAI reported a loss of $5 billion, with revenues of only $3.7 billion. It is estimated that OpenAI may not become profitable until 2029.
The precise definition of AGI has been a topic of much discussion. When OpenAI released its powerful o3 series model, there was speculation about whether it had already met the AGI criteria. The o3 model scored nearly 90% on the ARC-AGI benchmark test, surpassing human performance. However, the creator of the benchmark, Francois Chollet, stated on social media that he did not consider this to be AGI, as there were still simple ARC-AGI-1 tasks that the o3 model could not solve.
At the 2024 FinRegLab AI Conference, OpenAI CEO Sam Altman shared his views on the definition of AGI. He suggested that as we get closer to achieving AGI, the term becomes less useful. Different people have different interpretations of AGI, with some seeing it as true superintelligence and others as a form of intelligence that surpasses all of humanity. There was a time when people thought that OpenAI's o1 model was showing early signs of AGI.
For Microsoft, AGI is more about commercial benefits. Microsoft CEO Satya Nadella is unlikely to allow Sam Altman to declare the achievement of AGI based on a subjective definition. This has led to some strain in their relationship. To reduce dependency on OpenAI and lower costs, Microsoft plans to incorporate more internal and third-party AI models into Microsoft 365 Copilot. This move aims to diversify the underlying AI models, enhancing system stability and reliability.
Additionally, Microsoft may soon invest in another AI company, Anthropic, which developed the Claude model. According to a prediction by Dylan Patel, founder and chief analyst of SemiAnalysis, this new funding round could increase Anthropic's valuation to $59 billion. This provides Microsoft with new options in the AI landscape.
In summary, the consensus on the AGI definition between Microsoft and OpenAI lays the groundwork for their future cooperation, but their relationship also faces challenges. As AI technology continues to advance, balancing technological breakthroughs with commercial interests will be a key issue for both parties to address.
