In a recent interview, Yann LeCun, Chief AI Scientist at Meta, stated that we will soon have the technology to provide personal AI assistants, creating a team of personal digital assistants for everyone who can access the internet. As a scientist, he focuses on predicting technological advancements and their potential, emphasizing that we will have AI that can match or even surpass human intelligence. Image source note: Image generated by AI, image licensed from service provider Midjourney. LeCun pointed out that in the future, our relationship with AI will evolve.

Andrej Karpathy suggested that Reinforcement Learning from Human Feedback (RLHF) may not be the ultimate solution for AI to achieve human-level problem-solving capabilities. Using AlphaGo as an example, he pointed out that true reinforcement learning techniques optimize neural networks through self-play, ultimately surpassing humans without human intervention. In contrast, RLHF resembles imitation of human preferences rather than solving problems, being effective in closed environments with clearly defined reward mechanisms, such as Go, but less so in open-ended tasks like article summarization and code rewriting.

In the world of AI, we have just welcomed a remarkable new member—Cambrian-1, a large multimodal language model (MLLM) developed by industry giants LeCun and Xie Saineng. The emergence of this model represents not only a technological leap but also a profound reflection on the research of multimodal learning.

OpenAI's chief scientist LeCun has proposed a seven-stage development roadmap towards AGI. The 'Moonshot' plan aims to tackle the alignment problem of super AI, with recent research showing significant progress using small models to supervise larger ones. LeCun emphasizes that the first stage is to build systems that can learn how the world operates like small animals. He mocks the intelligence of current large models, stressing that the key to AGI lies in establishing brain-like world models. LeCun's vision includes creating goal-driven, safeguarded systems to gradually achieve super artificial intelligence.

LeCun believes that there will be intelligence exceeding human capabilities in the future, but emphasizes that it will not dominate humanity but rather act as an interactive mediator. Artificial intelligence must be open source to avoid corporate monopolies and ensure that everyone can participate in learning and optimization. There is no inherent connection between intelligence and domination; intelligent systems can be designed to be smart without the desire to dominate. LeCun calls for the establishment of a world model that predicts the consequences of action sequences to enhance general planning capabilities. Open source is key to the development of artificial intelligence, and LeCun is concerned about future challenges and hopes to draw wide attention.

The Galactica model from Meta was taken offline just three days after its release due to hallucination issues, with LeCun expressing regret. Galactica is regarded as an early attempt at large language models, positively influencing Meta's subsequent projects. The model's lead author, Ross Taylor, admitted that the foundational model was not thoroughly checked prior to release, but he still emphasized its research value. Joelle Pineau, Vice President of AI Research at Meta, highlighted that Galactica was a research project, not a product. The removal of Galactica has sparked significant discussion.

The AI regulatory debate is intense, with Andrew Ng supporting open regulation and advocating against excessive government intervention. LeCun and other leaders express concerns that strong regulation may lead to monopolies, emphasizing the negative impacts of regulation. The argument revolves around the threats of AI to humanity's future and the necessity of mandatory regulations. The U.S. government has issued an executive order on AI technology but has not provided specific protections for unemployed workers, highlighting the complexity of AI regulation and the trade-off between technology and social protections.

LeCun and xAI co-founder engage in a heated debate over GPT-4's reasoning abilities. LeCun argues that GPT-4 lacks reasoning skills, sparking widespread discussion. Christian Szegedy, co-founder of xAI, mentions the limitations and breakthroughs in reasoning abilities. Some netizens compare GPT-4 to human thinking, describing humans as 'random parrots.' A paper by Subbarao Kambhampati from ASU points out that GPT-4 performs poorly in planning reasoning.

Yann LeCun defended the Llama open source large model in the Senate, emphasizing the importance of open source. Artificial intelligence should be viewed as infrastructure, and open source can facilitate its rapid development. Open source models help democratize access to cutting-edge technology and reduce social disparities. There is an urgent need for collaboration to establish standards and governance models for artificial intelligence. Yann LeCun supports related legislation and is willing to cooperate with relevant personnel within companies.