DeepSeek LLM's 67B Base version has demonstrated superior performance compared to the Llama2 70B Base, particularly in areas like reasoning, coding, math, and Chinese comprehension.
The pilot includes testing over 15 Aidge APIs with software partners. These span key commerce areas like consumer insights, content localization, marketing, design, and customer service.
The brainchild of renowned computer scientist Kai-Fu Lee, 01.AI aims to develop AI systems tailored for the Chinese market.
The new model demonstrates remarkable capabilities in understanding complex instructions, copywriting, reasoning, memorizing, and preventing hallucinations.
The model can process around 350,000 Chinese characters at once. This vastly exceeds the previous benchmark of 75,000 English words set by Anthropic's Claude2 model earlier this year.
The substantial funding from Chinese tech stalwarts reflects their eagerness to cultivate domestic alternatives at a time when advanced US AI firms are increasingly out of reach.
The rules expand on export controls first imposed in October 2022 and represent the latest move in an escalating tech standoff between the two superpowers.
The fresh influx of capital sets a new record for the fastest rise to unicorn status among Chinese startups.
By emphasizing corpus safety, model security, and rigorous assessment, the regulation intends to ensure that the rise of AI in China is both innovative and secure—all while upholding its socialist principles.
The new models represent a significant step forward for multimodal AI that can process both visual and textual data in English and Chinese.
The ERNIE Bot plugins, combined with the release of an array of innovative AI solutions underscore Baidu's pole position in pioneering AI advancement in China.
Baidu, ByteDance, Tencent and Alibaba are aggressively stockpiling GPUs in light of tighter U.S. export controls.
With the new rules, China seeks to strike a balance between advancing their AI capabilities while maintaining rigid ideological control.
The model has set a new benchmark, outstripping Meta's LLaMa-13B by a substantial 40% in training data volume.