Tag
Forbes reports that US AI data vendors Surge AI, Mercor, AfterQuery, and Turing are selling ~$500M/year in training data to Chinese labs Tencent, Alibaba, and ByteDance, despite also serving OpenAI and Anthropic.
Seedance 2.5 introduces 30-second video generation, multimodal references, multilingual creation, and targeted editing, with API access coming soon on BytePlus for developers and enterprises.
ByteDance is training a massive AI model, aiming to rival Anthropic, with an independent development approach led by its Seed team under former Google DeepMind scientist Wu Yonghui, while its Doubao model leads China with 324 million MAU.
ByteDance is in the early stages of training a large language model with up to 10 trillion parameters, signaling a massive scale-up in AI development.
据金融时报报道，字节跳动正在训练一个估算参数量达10万亿级别的AI大模型，规模接近Anthropic的先进系统，旨在缩小与美国顶级AI实验室的差距。
ByteDance is expanding its AI education app Gauth with AI-generated animations to walk students through problem-solving, raising questions about whether such tools genuinely improve comprehension or merely create an illusion of learning.
ByteDance announced SeedRealtime, a native audio-visual model that can process continuous video, audio, and text while speaking in real time.
This tweet argues that Zhang Yiming is not simply an engineer who loves technology, but a master of office politics who secured his position through a series of power plays, citing the development of Douyin and Volcano as examples.
ByteDance is integrating its Seedance video generation model (v2.5) into its study app Gauth to produce AI-narrated animated lessons with quizzes, starting Aug 4. The move marks one of the first large-scale deployments of generative video for education and raises questions about fact-checking and copyright.
ByteDance Seed officially launches Seedance 2.5, a new-generation video creation model with 30-second single-pass generation, multi-round extensions, multimodal referencing, and precise editing capabilities, now rolling out on Jimeng AI and Doubao Pro.
Neill Blomkamp released a 13-minute sci-fi short 'Nightborne' made entirely with ByteDance's Seedance 2.0 text-to-video AI. The Verge criticizes it as slop, noting the AI's limitations and lack of creative voice.
ByteDance debuts Seedream 5.0 Pro, a multimodal image-creation model for production design with advanced reasoning, complex visualization, precise editing, realistic textures, and native multilingual support.
ByteDance launches Seedream 5.0 Pro, a multimodal image creation model with breakthroughs in complex information visualization, interactive precision editing, realistic imagery, and native multilingual support.
ByteDance is set to launch Seedance 2.5, an AI video model capable of generating up to 3-minute videos, with features like 30-second scenes, multimodal references, and integration into Dreamina and CapCut.
ByteDance publicly shared internal data at the FORCE Conference: the TRAE team generates over 90% of its code via AI, but per capita requirement throughput increased by only about 60%. This shows that while code generation is fast, downstream engineering processes become the bottleneck, and AI programming is entering its DevOps moment.
ByteDance's Seed 2.1 model achieved strong results on multimodal agentic (Claw-Eval) and long video understanding (Video-MME) benchmarks, though a gap remains between perception and agentic capabilities.
ByteDance unveiled Seedance 2.5, an AI video model that can generate 30-second 4K clips from a single prompt or up to 50 reference pieces, rivaling OpenAI and Google's offerings.
ByteDance has released Seed2.1, a new AI model, with accompanying blog post and model card.
Discusses ByteDance's rich product ecosystem and emerging new products, with links to development documents of Coze, Volcano Engine, and Feishu.
Stanford and ByteDance introduce W-Flow, a single-step generative model that uses Wasserstein gradient flows to achieve state-of-the-art one-step ImageNet 256x256 generation (1.29 FID) with 100x faster sampling than multi-step diffusion models.