ByteDance reportedly readies a Muse-like AI assistant for release within weeks
The reported product push comes as co-founder Zhang Yiming puts competition with OpenAI and Anthropic at the research frontier years away.
Loading page…
The reported product push comes as co-founder Zhang Yiming puts competition with OpenAI and Anthropic at the research frontier years away.
Listen to this story
ByteDance is reportedly aiming to release a Muse-style assistant within weeks, while co-founder Zhang Yiming told AI staff that matching OpenAI and Anthropic at the research frontier could take until 2029 or 2030. Zhang reportedly pointed to a substantial training-data gap: ByteDance’s Seed models use pretraining datasets orders of magnitude smaller than those of frontier competitors. The product push therefore appears to have a nearer-term timetable than ByteDance’s bid to catch up in advanced model research; the assistant’s release date and any hardware plans remain unconfirmed.
Meta introduced Muse on September 8 as an agent that can work across connected apps, use a browser, fill out forms, and take instructions through WhatsApp.
Zhang’s 2029-or-2030 estimate concerned competing at the research frontier, not when ByteDance could ship an assistant.
Bloomberg reported the planned release; no exact date is confirmed, and ByteDance did not respond to its request for comment, according to The Next Web.
ByteDance is preparing an AI assistant similar to Meta’s Muse for release in the coming weeks, Bloomberg reported on November 11. The near-term product push comes with a longer research horizon: co-founder Zhang Yiming reportedly told staff that competing with OpenAI and Anthropic at the cutting edge could take until 2029 or 2030.
The report draws on people familiar with a recent all-hands meeting of ByteDance’s AI team.
Meta introduced Muse on September 8, describing a personal AI agent that takes on tasks and projects rather than simply answering questions.
According to Meta, Muse can work across apps a person connects, open a browser and fill out forms. People can give it instructions through its own app or WhatsApp. The company also says it can turn a longer-term goal into a plan and advance the work without requiring a new instruction for every step.
Zhang also reportedly identified a gap in training data. ByteDance’s Seed language models are pretrained on datasets orders of magnitude smaller than those of its frontier competitors, he told staff. Pretraining is the model’s initial training stage; his comparison concerns the scale of the material used there, rather than a measured result for the planned assistant.
Whether ByteDance will pair its assistant with hardware is unclear.
A ByteDance spokesperson did not respond to Bloomberg’s request for comment, according to The Next Web. No exact release date is confirmed.
Loading discussion...
Join the conversation
Explain how much model leadership would influence your choice.
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.