The co-founder of OpenAI was interviewed by Huang Renxun: GPT-4 's reasoning ability has not met expectations.

CTOnews.com March 23 news, recently, Nvidia CEO Huang Renxun and OpenAI co-founder and chief scientist Ilya Sutskever had an in-depth conversation. According to CTOnews.com, the conversation was recorded a day after the release of GPT-4, OpenAI's most powerful artificial intelligence model to date, and they discussed in detail GPT-4 and its predecessor, including ChatGPT.

Sutskever says GPT-4 is "quite improved in many ways" compared to ChatGPT, noting that the new model can read images and text. "in a future version, [users] may get a chart as an answer," he said. "

When asked if GPT-4 has reasoning ability, Sutskever said that the term is difficult to define, and this ability may still be in the pipeline, GPT4 has not yet reached the level of logical reasoning, the team will continue to improve their modeling skills, but there is no guarantee of how much reasoning can be achieved. "We believe that GPT's reasoning ability has not reached the expected level, and I am confident that if we further expand the database and maintain the business operation model, its reasoning ability will be further improved," he said.

Sutskever also predicts the future of generative AI. "at present, we just provide text to AI and let it draw conclusions, but AI cannot verify the authenticity of the text and its source. The next step is for the neural network to verify the authenticity of the data source and make it aware of the needs of users."