OpenAI and competitors seek new path to smarter AI

Leading AI companies including OpenAI are developing new training techniques as current methods reach their limits. The goal is to construct more human-like ways for algorithms to “think.”

What you should know:

• OpenAI and its competitors are encountering unexpected delays in LLM development due to current training techniques reaching their peak.?
• Reuters says a dozen AI researchers believe that new techniques, such as those behind OpenAI’s recently released o1 model, could reshape the race toward AGI.?
• Major AI companies previously maintained that “scaling up” existing models by adding more data and computing power would consistently lead to improved models.?
• Now that the effects of scaling up pre-training data have plateaued, researchers are turning to a “test-time compute” technique that can enable AI models to plan their responses more thoroughly rather than providing immediate answers to user queries.

What this means: The AI sector is finally moving past its longstanding “bigger is better” philosophy. OpenAI, Anthropic, xAI and Google DeepMind are all developing their own versions of the test-time compute technique. While this may delay the release of the next generation of LLMs, it also suggests that they will be far more advanced than the models we have today.?