IT House reported on November 18 that there are recent rumors that top AI laboratories including OpenAI, Google and Anthropic are encountering difficulties in developing advanced AI models. These reports attribute the cause to a lack of quality training data, leading to diminishing returns that make continued investment difficult.
In this regard, OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei both expressed different opinions. Altman appeared to respond to the negative press by posting a cryptic message on social media suggesting "there are no limits." Amodei shares a similar view, and while he acknowledges that many factors may hinder the development of AI, he is optimistic that there are ways to overcome these obstacles, including the use of synthetic data.
IT House noticed that former Google CEO Eric Schmidt also expressed his views on these reports. He believes that large language models will undergo two to three iterations in the next five years . The performance of these models is increasing at an unprecedented rate, and there is currently no evidence that this growth trend is about to stop . Although we will eventually hit a bottleneck, we are not there yet.
Schmidt predicts that more advanced and powerful AI models will emerge in the future, especially in the fields of physics and mathematics. He explained OpenAI's recently released Strawberry AI model OpenAI-o1 as an example. However, Schmidt also warned that as models become more advanced, the potential dangers increase. He noted that these primitive models may be better at launching "zero-day attacks" than humans . In addition, biological viruses and new types of warfare may also become potential threats brought by the development of AI.