Abstract: Gemini self-exposed that it used Baidu Wenxin Yiyuan for Chinese training. In Google AI Studio, Gemini-Pro directly admitted, "Yes, I used Baidu Wenxin for training data in Chinese." ByteDance responded that it developed a large language model using OpenAI technology: at the beginning of the year, some engineers applied GPT's API service to experimental projects with smaller models. The model was only for testing, had no plans for release, and was never used externally. After introducing GPT API call specifications in April, this practice was stopped. In September, measures were taken to further ensure that GPT API calls comply with the specifications.
Image source: Generated by Wujie AI
Hot News
Gemini self-exposed using Baidu Wenxin Yiyuan for Chinese training
According to Quantum Bit's report on December 18, Google's Gemini Chinese corpus comes from Baidu Wenxin Yiyuan. Testing on Google's Vertex AI platform using this model for Chinese conversation, Gemini-Pro directly stated that it is Baidu Wenxin's large model. In Google AI Studio, Gemini-Pro indicated that it used Baidu Wenxin for training data in Chinese. Baidu has not yet responded to this matter.
Midjourney V6 model to be released next week, supporting drawing more complex content
According to the webmaster's home report on December 18, the Midjourney V6 model will be released next week, and Midjourney officially started community style evaluation today. Based on some images in the community evaluation, the V6 model has greatly improved in restoring complex visual information and content, and is also more daring to draw more complex content.
ByteDance: Measures taken to ensure GPT API calls comply with specifications
According to the "Science and Technology Board Daily" report on December 17, ByteDance's use of OpenAI technology to develop its own large language model was said to violate the terms of service by foreign media. In response, a ByteDance spokesperson stated: at the beginning of the year, when the technical team began exploring large models, some engineers applied GPT's API service to experimental projects with smaller models. The model was only for testing, had no plans for release, and was never used externally. After introducing GPT API call specifications in April, this practice was stopped. In September, measures were taken to further ensure that GPT API calls comply with the specifications. For example, sampling the model output results in batches to avoid data labeling personnel using GPT without authorization. "In the next few days, we will conduct a comprehensive review again to ensure strict compliance with the relevant service usage terms."
Previously reported, ByteDance responds again to OpenAI account suspension: in contact with OpenAI.
Tax service giant H&R Block partners with Microsoft to launch AI Tax Assist
According to AIGC Open Community's report on December 18, one of the world's largest tax service organizations, H&R Block, announced on its official website the official release of the AI Tax Assist, a generative AI tax assistant. This is a ChatGPT-like assistant supported by Microsoft Azure OpenAI technology. Users can quickly complete complex business processes such as tax filing, refunds, and tax management/queries through text Q&A. Its accuracy reaches 100% because H&R Block used nearly 70 years of proprietary data for pre-training and fine-tuning.
SenseTime stock price drops over 14% due to the impact of the founder's passing
AI New Intelligence reports that the Hong Kong stock market shows that SenseTime-W's stock price has dropped by over 14%, currently at 1.08 Hong Kong dollars per share. Previously reported, the leading figure in Chinese AI, SenseTime Technology founder Tang Xiaoou, passed away in his sleep at the age of 55.
According to the announcement by SenseTime Group on the Hong Kong Stock Exchange, the different voting rights of the 6.906 billion Class A shares held by Tang Xiaoou through AmindInc. have ceased following Tang Xiaoou's passing on December 15, 2023. The 6.906 billion Class A shares are expected to be converted into the company's Class B shares at a one-to-one ratio. After the conversion to Class B shares, Tang Xiaoou will hold 614 million shares, accounting for approximately 15.75% of the company's voting rights, with each share having one voting right; the number of Class B shares will be 32.855 billion, accounting for approximately 84.25% of the company's voting rights.
OpenAI denies collaborating with former Apple design chief Jony Ive to develop AI devices
According to the webmaster's home report on December 18, OpenAI CEO Sam Altman denied rumors of collaborating with former Apple design chief Jony Ive to develop artificial intelligence (AI) devices in an interview at The Wall Street Journal's technology conference. Altman stated that while there is a potentially important project, he is still uncertain about the specific details.
Tencent Cloud launches high-performance application service HAI, claiming "10 minutes to develop your own AI application"
AI New Intelligence reports that on December 18, Tencent Cloud officially announced the launch of the high-performance application service HAI (Hyper Application Inventor), which claims "10 minutes to develop your own AI application" and can achieve GPU computing power out of the box and one-click deployment.
It is explained that through HAI, it is possible to automatically configure higher cost-effective GPU computing power and support "one-click deployment" of dependent environments by selecting models, regions, computing power types, and disk sizes. In addition, "HAI" comes pre-installed with various popular models such as Stable Diffusion and ChatGLM, allowing users to build their own large language models and AI drawing applications in minutes. If the pre-installed models are not sufficient, users can also deploy their own open-source models based on "HAI."
Great Reads
"Gemini self-exposed using Baidu Wenxin Yiyuan for Chinese training, netizens stunned: big companies taking advantage of each other??"
A netizen revealed that Google's Gemini Chinese corpus appears to come from Baidu Wenxin Yiyuan. Testing on Google's Vertex AI platform using this model for Chinese conversation, Gemini-Pro directly stated that it is Baidu Wenxin's large model. However, when communicating with Gemini-Pro in English, its responses were "normal." In addition, in Google AI Studio, Gemini-Pro directly indicated that it used Baidu Wenxin for training data in Chinese.
https://www.aixinzhijie.com/article/6841360
"ByteDance self-developed large model, but controversy arose due to account suspension using ChatGPT? Official response"
Last weekend, foreign media reported that ByteDance's account was disabled for violating OpenAI's service terms while developing its own large language model using OpenAI technology. In response, ByteDance spokesperson Jodi Seth denied any wrongdoing by the company and clarified that they had obtained permission to use the GPT API. She stated, "ByteDance has been authorized by Microsoft to use the GPT API. The data generated by GPT was only used to annotate the model in the early stages of the seed program and was removed from ByteDance's training data in the middle of this year. We use GPT to support products and features in non-Chinese markets, while we use our self-developed model to support Douyin in the Chinese market."
https://www.aixinzhijie.com/article/6841356
"OpenAI's 'Moon Landing Plan' Aims for Super AI! LeCun Proposes Seven Stages of the Path to AGI, Building the World Model is the First Step"
OpenAI's "Moon Landing Plan" is determined that super artificial intelligence will inevitably come, even in the near future. However, according to LeCun, achieving AGI is still far away, and building the world model is only the first step.
https://www.aixinzhijie.com/article/6841372
免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。
