[Zhang Yiming: Seed's Training Data is Several Orders of Scale Smaller; ByteDance Won't Be Able to Compete with OpenAI and Anthropic Until 2029 at the Earliest] Beating AI News reports that Zhang Yiming recently gave a timeline at a Seed team meeting: ByteDance won't be able to compete with OpenAI and Anthropic in cutting-edge AI fields until 2029-2030 at the earliest. Zhang Yiming specifically pointed out that the pre-training dataset used by Seed's large language model is several orders of magnitude smaller than that of its top competitors. ByteDance has already acquired many strong positions in the AI field. Doubao is the largest AI chat application in China in terms of user base, and Seedance is among the world's top video generation models. However, its general language model has lagged behind. When Seed 2.1 was released in June this year, the team itself admitted that the model still has significant shortcomings in high-difficulty open tasks and cutting-edge research problems. ByteDance is prepared to invest a massive amount of money in this area. In May, the company discussed increasing its full-year capital expenditure to a maximum of $70 billion, primarily for building AI data centers and other infrastructure. In August, it was reported that Seed was pre-training a massive model with up to 10 trillion parameters, exceeding the scale of all publicly available large models in China. Zhang Yiming also doesn't want to take shortcuts through distillation. Distillation involves using answers generated by leading models like Claude to train one's own model. In August, he explicitly stated that while this method can improve performance in the short term, what is learned is ultimately existing capabilities; at best, one can only approximate them, but it's difficult to truly surpass them. Internally, Seed also restricts the input of third-party models into the training set through API checks and training data filtering. To catch up, ByteDance is also constantly recruiting and retaining talent. Seed currently has approximately 1,000 R&D personnel, led by Wu Yonghui, whom Zhang Yiming poached from Google DeepMind last year. ByteDance has also launched virtual shares called "Doubao Shares" specifically linked to its Doubao business. Eligible employees can choose to receive their year-end bonuses as Doubao Shares, sharing in the profits generated by business growth. ByteDance plans to launch a personal AI assistant similar to Meta Muse in the coming weeks, allowing users to interact with AI through voice and virtual avatars and have it complete tasks on their behalf. The Doubao personal assistant project, internally codenamed "Spell," began internal testing as early as April, but was delayed due to factors such as the release of the Doubao phone. After the success of Meta Muse, ByteDance accelerated its progress, working on development before the National Day holiday, planning to further integrate Spell with Doubao. [Original Link]