Former ByteDance Seed engineer: ByteDance iteration takes half a year, Google rumor says only three months.

ME News, April 24 (UTC+8): According to Beating’s monitoring, Zhang Chi, a former engineer from ByteDance’s Seed team and now an assistant professor at Peking University, revealed on the podcast Into Asia that it takes ByteDance about half a year to complete a round of large-model training (pre-training plus post-training), while Google reportedly only needs three months. He believes that iteration speed is one of the core reasons Chinese companies struggle to catch up. Zhang Chi spent about a year at ByteDance, and his mathematics team was more research-oriented. He said that the team’s positioning was “more for publicity,” unlike the pre-training and post-training teams responsible for model delivery.

Zhang Chi described Seed’s internal “benchmaxxing” culture (score-chasing): team leaders evaluate performance based on the benchmarks they are responsible for, and everyone is trying to rack up more scores, “but this cannot translate into a good experience in real-world use.” He said that on paper, the models from major Chinese companies can keep pace with leading U.S. models, but in practice, they are “not good enough.” Seed’s goal is to be among the world’s top, “but unfortunately, I don’t think we have caught up,” and even the goal of being number one domestically “was not achieved.”

At the end of 2024, Seed claimed that it had caught up with GPT-4o, but then DeepSeek was released, and the team realized the gap still remained. When he joined, the entire team was urgently pivoting to reinforcement learning. (Source: BlockBeats)