OpenAI CEO Sam Altman has argued that a significant portion of the research community has held back AI development by underestimating the power of scaling large language models. Altman, speaking at Stanford, criticized the field for being overly confident about the limits of scaling, which he believes has stifled innovation. He emphasized that betting against the potential of large language models at this stage is misguided, given the recent advancements in the field.
Altman highlighted that models like those developed by OpenAI have already surpassed human intelligence in certain areas, such as disproving a long-standing mathematical conjecture. This achievement has sparked discussions within the academic community, with mathematicians questioning how such models could reshape their field. He also noted that while world models are crucial for applications like robotics, the data clearly supports continued scaling as the path forward.
According to the source, Altman addressed critics like Yann LeCun, who has previously dismissed large language models as a dead end. He argued that some researchers are too attached to their positions to accept data that contradicts their views. Altman also dismissed the influence of 'Twitter trolls' who have long predicted OpenAI's failure, stating that their predictions have not materialized.
Source: thedecoder
