OpenAI and competitors explore new training techniques as AI model scaling reaches limits

AI companies, including OpenAI, are shifting focus from simply increasing the size of models to developing new training techniques that mimic human thinking. This change comes as researchers face challenges with current methods, which have shown diminishing returns. Ilya Sutskever, co-founder of OpenAI, noted that the era of merely scaling up models is over. His new company, Safe Superintelligence, is exploring alternative approaches to improve AI performance. Meanwhile, OpenAI's latest model, o1, uses a technique that enhances decision-making during its operation. This shift may impact the demand for AI hardware, particularly Nvidia's chips, which have dominated the market. Investors are observing how these changes could alter the competitive landscape in AI development and hardware supply.