On the morning of January 10th, Beijing time, Intel CEO Pat Gelsinger delivered a keynote speech at the US International Consumer Electronics Show (CES). As an important part of Intel's exhibition theme event - "AI is everywhere, starting with Intel", this speech is undoubtedly one of the most watched keynote speeches at this year's CES exhibition.

At the event, Gelsinger spoke with CNBC reporter Kristina Partsinevelos for nearly half an hour. During the conversation, while Gelsinger didn't make a major product announcement, he took a higher-level look at product strategy and the future of computing, and discussed a lot about artificial intelligence in depth.

The world is discovering the value of artificial intelligence

When it comes to artificial intelligence, when the outside world discusses its potential, it often also talks about the potential risks it may bring. On this issue, Gelsinger classifies himself as an optimist, emphasizing that “Artificial intelligence can be shaped into an incredible force for good."

He believes that if AI risks are to be reduced, "open access to technology is a critical part so people can see how AI works."

Gelsinger is obviously confident about the future prospects of artificial intelligence. He said that he does not believe that the scientific development of artificial intelligence is approaching the end. On the contrary, the current stage of AI development is "similar to the early days of PC development or the early days of Moore's Law."

Gelsinger said:

"The world is being driven to discover the possibilities that AI brings to productivity. The world will discover the value of AI in various fields, and new applications of AI will have new economic impacts."

A legal framework related to AI needs to be established

Gelsinger also talked about the issue of AI regulation in his speech.

Recently, many artificial intelligence companies, including OpenAI and others, have faced copyright-related legal proceedings. The lawsuits allege that companies such as OpenAI infringe the copyrights of the original authors by copying the plaintiffs' works on a large scale without permission or consideration and then feeding the plaintiffs' copyrighted works into their large language models (LLMs) for training.

Gelsinger pointed out that the training process of LLM requires a large amount of content, whether it is authorized content or unauthorized content. A trained LLM is basically equivalent to compressing the Internet into a model. Therefore, establishing a legal framework around this issue is inevitable.

"Artificial intelligence is developing rapidly. But politicians are slow to move." Gelsinger believes that these cases will eventually go to court, thus providing a legal precedent for the latter.

Intel AI accelerator performs 'very well'

Gelsinger also revealed news about the third-generation Intel AI accelerator Gaudi3 during his speech.

In December last year, Intel debuted its Gaudi3 series of AI accelerators for deep learning and large-scale generation of artificial intelligence models. Intel claims that it plans to officially release Gaudi3 in 2024, and claims that the performance of Gaudi3 will be better than Nvidia’s flagship AI chip H100.

In remarks on Tuesday, Gelsinger revealed,Gaudi3 has now shipped from the factory and is in the early testing and debugging stage. Machine manufacturers and corporate customers will be involved in the commissioning process in the coming months.

He also said that the chip's performance so far looks "very good." Later this year, the company will officially release this blockbuster new product.