Exploring the boundaries of reasoning abilities
Researching intrinsic rewards and proactiveness in models, memory mechanisms, and online learning capabilities.


The Seed STEM Fellows Program invites 100 outstanding researchers from the forefront of scientific innovation to collaborate with the Seed team on real-world research challenges. Together, we aim to accelerate scientific discovery with AI and advance the frontier of AI intelligence.
The Seed STEM Fellows Program invites 100 outstanding researchers from the forefront of scientific innovation to collaborate with the Seed team on real-world research challenges. Together, we aim to accelerate scientific discovery with AI and advance the frontier of AI intelligence.
Researching intrinsic rewards and proactiveness in models, memory mechanisms, and online learning capabilities.
Studying how to scalably extract knowledge from massive, ultra-low signal-to-noise ratio multimodal data and unify multimodal generation and understanding.
Exploring hardware-friendly model architectures and training methods and developing models with long-term memory and real-time interaction.
Discovering how to take actions based on visual and speech inputs and utilize various tools in the process flexibly.
Agents exhibit a clear log-sigmoid pattern in environment learning.
Enhancing latent reasoning through looped language models.
Possess deep curiosity and a passion for fundamental research.
Have top-tier research capabilities, dare to dream big, and pursue ambitious, high-impact AI challenges.
Stay resilient and persistent amid short-term changes, and remain committed to long-term research.
Diverse research backgrounds, with small and flat research groups where you can easily propose new topics.
Extended evaluation cycles with no OKRs or semi-annual reviews, focusing on long-term impact.
Ample computing resources allocated and managed per research project.
We encourage cross-team and cross-modal communication and collaboration to promote the interconnection and integration of cross-modal knowledge.
If you have any questions,
feel free to reach out via:
seededge@bytedance.com