Recent AI News

1. OpenAI officially releases GPT-4

At 2 a.m. on March 15, OpenAI officially launched GPT-4, which has significantly improved in logic ability, language ability, and factual accuracy compared with the previous GPT-3.5. GPT-4 supports polymorphic input, including images and text, and can extract and understand text from images. ChatGPT based on GPT-4 has a higher sense of humanity, safety and morality, and its language capabilities are more abundant. In the future, we have reason to expect that many excellent products based on GPT-4 will emerge.

Currently, GPT-4 is only open to the development team and ChatGPT Plus users to apply for the Waitlist. In terms of price, GPT-4 API is priced at $0.03 per 1,000 prompt tokens and $0.06 per 1,000 completion tokens. The default rate limit is 40,000 tokens per minute and 200 requests per minute. The context length of GPT-4 is 8,192 tokens, which is slightly slower than GPT-3.5 Turbo. For more detailed evaluations of GPT-4, please read: Understand the release of GPT-4 in a few minutes

2. Microsoft announced the launch of Microsoft 365 Copilot, reshaping productivity through AI

Microsoft is bringing the large language models used by ChatGPT to its Office suite. The software giant announced Microsoft 365 Copilot, similar to GitHub Copilot, uses AI to help office software users improve productivity. 365 Copilot can generate documents and presentations according to user requirements, analyze Excel data, summarize Outlook emails, summarize discussion points in Teams, automate repetitive tasks, and more. Microsoft will make Copilot available to Microsoft 365 users in the coming months, and the software giant hasn't disclosed whether it will cost extra.

3. Visual ChatGPT based on ChatGPT and visual model

Recently, Microsoft released Visual ChatGPT on GitHub, which connects ChatGPT and a series of visual basic models, allowing users to send and receive pictures during chat.

Visual ChatGPT Demonstration In the actual use process, we found that Visual ChatGPT has a strong performance in image recognition, but it still needs to be strengthened in generating images. In addition, in the actual demonstration, we found that Visual ChatGPT has a high requirement for video memory. Even with 12GB of video memory, the system only supports the use of three models, so the effect is limited. We don't know if we can improve image generation quality if we use more models.

Hugging Face official demo: address

GitHub project: address

4. Google releases PaLM

Google recently released the PaLM model. For more details, see the original link .

Google released PaLM. Compared with OpenAI's launch of GPT-4, Google's announcement of the release of the PaLM API does not seem to have received much attention. PaLM can complete tasks such as text and code generation and editing. At the same time, Google also launched a tool called MakerSuite to help developers make better use of these technologies. In addition, Google also announced that it will apply AI functions to applications such as Gmail and Docs, and provide more AI functions, such as optimizing email and slideshow production. As long as the subject is entered, AI can help us quickly generate articles or emails and improve work efficiency. However, PaLM is currently not open to ordinary users, and only some testers can try it out. At the same time, Microsoft also demonstrated a similar technology on March 16, integrating AI technology into the daily work environment.

5. Baidu launches AI product "Wen Xin Yi Yan"

Robin Li, founder, chairman, and CEO of Baidu, announced the official release of "Wen Xin Yi Yan" on the afternoon of March 16. It is a generative dialogue product developed based on Wenxin large model technology, similar to ChatGPT. In addition, "Wen Xin Yi Yan", also known as ERNIE Bot, has cross-modal and cross-language in-depth semantic understanding and generation capabilities. It is expected to be widely used in fields such as search questions and answers, content creation and generation, and smart office.

After the press conference, Baidu only opened the invitation channel, and the detailed evaluation will appear in the follow-up content. However, it should be noted that Baidu’s stock price on the Hong Kong Stock Exchange (09888.HK) fell by nearly 8.7% within just one hour of the press conference. Judging from the market reaction, this product did not meet the expectations of the public.

6. Coca-Cola’s latest ad uses Stable Diffusion AI

On March 6, Coca-Cola used Stable Diffusion AI technology and 3D technology in its latest advertising video for advertising shooting. This technique produces many stunning artistic effects.