Last November, what was hailed as artificial intelligence Yann LeCun, the French scientist behind The Godfather, suddenly announced his departure from Meta, ending his role as the company's chief artificial intelligence scientist.
More than a month later, LeCun, who had just started his own business, revealed in an interview that he had a lot of opinions about Meta.
He revealed that starting last year, Meta adopted a more aggressive strategy in the field of artificial intelligence . Meta CEO Mark Zuckerberg put more pressure on the generative AI division to accelerate research and development and deployment, which led to poor internal communication.
Yang Li-kun said the team had many new ideas and great ideas, but Meta only chose those that were safe and proven to work, which he felt meant falling behind.
Subsequent Llama models did indeed perform poorly. Yang Likun also admitted that the Llama 4 model released in April 2025 exhibited fraudulent practices in benchmark tests: the team used different models in different benchmarks to obtain better test results.
Zuckerberg was also very angry about the poor results of Llama 4 and marginalized the entire generative AI department, leading to the departure of many people.
To make matters worse, Zuckerberg attempted to propel the team forward by introducing a "catfish." Last June, Meta invested $15 billion in data labeling startup Scale AI and hired its CEO, Alexandr Wang, to lead Meta's cutting-edge AI model development, meaning he became Yang Likun's superior.
However, Yang Likun felt that 28-year-old Alexandr Wang was too young and inexperienced. Yang Likun said that Wang learned quickly and was aware of his shortcomings, but he had no research experience, did not know how to conduct research activities, and was unaware of the likes and dislikes of researchers.
Yang Li-kun bluntly stated that continuing to work at Meta has become administratively difficult because the company's new round of recruitment for superintelligence projects is focused on Large Language Models (LLM). He also revealed that many people at Meta do not want him to make public statements, as Large Language Models are essentially a dead end in the field of superintelligence.
Another reason that prompted him to leave Meta was that he explored other applications using world models and advanced machine intelligence, such as jet engines and heavy industry, but Meta was not interested in these.
Yang Likun emphasized that startups engaged in basic research will be the most promising new development area. This includes his new company, as well as the startup of former OpenAI CTO Murati and the founding team of OpenAI co-founder and former chief scientist Sutschkeville.
(Article source: CLS)
