ByteDance's founder says catching the AI frontier is years away, 2029 at the earliest
Zhang Yiming told ByteDance's AI team the company can compete with OpenAI and Anthropic, but will not match the leading edge before 2029.
The short version
- Zhang Yiming says ByteDance can compete with Anthropic and OpenAI, but not reach the frontier before 2029
- He reportedly said Seed's training data is several orders of magnitude smaller than leaders' data
- He told staff not to rely on distillation, meaning training on a bigger model's answers
- Bloomberg, via sources, says a Muse-like AI assistant could launch within weeks
- ByteDance's video model Seedance is the area where it is already seen as strong
Why it matters
A founder who runs one of the world's biggest consumer tech companies is saying the leading Western labs will stay ahead for years. For anyone choosing AI tools, the best-known options are unlikely to be displaced soon, but credible alternatives from China will keep growing, especially in video.
A sober timeline
As tmtpost.com reported, ByteDance co-founder Zhang Yiming spoke at a recent all-hands meeting of the company's AI team. His message: ByteDance can compete with Anthropic and OpenAI, but truly matching the frontier will take until 2029 at the earliest.
That is a notable contrast with Silicon Valley voices who talk about human-level AI arriving within a couple of years. Competing and catching up are different claims, and he is making the second one cautiously.
The data gap and the no-shortcut rule
According to the report, Zhang admitted that the dataset used to pretrain ByteDance's Seed language model is smaller than its top rivals' by several orders of magnitude. Pretraining is the first, large-scale stage where a model learns from text, so the amount and quality of data set a ceiling on how capable it can get.
He also reiterated that ByteDance should not depend on distillation: training a model on the outputs of a larger, stronger one. It is a known shortcut for climbing benchmark scores quickly, and peers including Anthropic's CEO Dario Amodei have criticized it. For a company that relies on TikTok abroad, a copyright or data-source dispute would be costly.
Popular app, quiet model
ByteDance's chatbot Doubao is described as the most popular in China, but the report argues that this comes from product skill, free access and traffic from Douyin, not from a model lead. Seed itself has kept a low profile: it has not topped global benchmarks or released frontier model weights to developers.
To close the gap, ByteDance hired Wu Yonghui from Google DeepMind to lead Seed and gave him priority equity.
Where ByteDance is already strong
The report says the Seedance video model is a real edge, helped by the huge video library behind Douyin and TikTok. It also notes that OpenAI scaled back its Sora video service, leaving room for Seedance.
Bloomberg, citing people familiar with the matter, says ByteDance plans to release an AI assistant like Meta's Muse within weeks. The report also says ByteDance set up a separate safety and alignment team, which helps it meet rules in different markets.
What to watch
- Does ByteDance actually launch the Muse-like assistant in the coming weeks, and what can it do?
- Will the next major Seed language model release narrow the gap Zhang describes?
Try it
When choosing an AI tool for a team or class, check whether you can export your data and switch providers, and re-test your choice every few months.
Sources and links
- Reportingtmtpost.com: the article we sourced
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.
Listen to the AI Wire headlines
Listen to the AI WireListen to our daily AI shows every morning: on your commute, in the shower, wherever you are
Listen to the daily showsFirst reported by tmtpost.com · Read the original. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.