AI points of the week: a wave of unrest, another wave, this may be the most "overcoming calamity" week for OpenAI
The past week has not been satisfactory for OpenAI. First, the networking function was outsourced, the website traffic declined, and then the author sued for copyright infringement. It can be said that the negative information is full. However, today, OpenAI announced that the use of the GPT-4 API will be released. It is worth noting whether it can reverse the trend.
A rough week for ChatGPT
Reversing the decline, OpenAI made a big move on Friday
OpenAI, which has been receiving negative information in the past few days, finally released an update today...
ChatGPT suspend networking function
On July 4th, OpenAI announced that it would suspend the ChatGPT web browsing function based on the Bing search engine. Officials stated that they noticed that this function sometimes displayed content in ways they did not want to see, and was frequently abused.
Foreign media speculate that this is because ChatGPT's networking function will bypass the paywall and privacy settings of some websites when searching the Internet, and will also tell ChatGPT users the complete non-public content.
Although OpenAI said that the networking function will "be back online soon," it did not give a specific timetable.
ChatGPT website traffic decline
According to SimilarWeb data, the growth of ChatGPT slowed down significantly in June, and the downloads of ChatGPT for iOS in the US fell by 38% month-on-month.
In terms of website traffic, global user traffic on the ChatGPT website and mobile terminal decreased by 9.7% month-on-month, the number of unique visitors decreased by 5.7%, and the time spent by visitors on the website also decreased by 8.5%.
Horror writer sues OpenAI
On July 5, OpenAI was involved in a copyright dispute over training data. Two well-known horror writers claimed that ChatGPT had illegally embezzled their data for training. The court is currently hearing the case.
Two writers, Paul Tremblay, author of The Cabin at the End of the World, and Mona Awad, author of Rabbit, felt that ChatGPT had written extremely detailed book briefs, which is enough to show that their novels were Used to "train" ChatGPT.
ChatGPT fully open GPT-4 API
On July 7th, OpenAI announced the full opening of the GPT-4 API with 8K tokens context to paying API users. All API developers with payment records can now use it without queuing, and new developers can get GPT-4 at the end of this month . 4 API Permissions.
In addition, OpenAI expects to open the fine-tuning function of GPT-4 and GPT-3.5 Turbo later this year. At the end of the blog post, OpenAI also disclosed the deprecation plan of a large number of old models and APIs. What's even more exciting is that OpenAI's official announcement: The code interpreter will be open to all ChatGPT Plus users next week .
Wenxin Yiyan launched on the App Store
After downloading, you still need to apply for internal test qualification
On July 3, "Wen Xin Yi Yan" was launched on the App Store. The installation package size is about 44.3MB, and it supports iPhones with iOS 13 and above.
It is worth mentioning that in early April, Baidu also sued Apple for the emergence of a large number of counterfeit "Wen Xin Yi Yan" apps on the App Store.
It should be noted that the current use of "Wen Xin Yi Yan" still needs to apply for internal test qualifications before it can be used.
Legal AI ChatLaw released
A free web version will continue to be available
On July 3, the Peking University team announced its newly developed ChatLaw, a large model in the legal field, which can provide services for answering questions and legal advice for legal novices.
The development team built a legal knowledge base based on a large number of original judgments and laws and regulations, and at the same time cooperated with law schools and well-known law firms to ensure that the knowledge base can be updated in a timely manner.
In the training phase, the development team performs special processing and enhancement of these knowledge to ensure that the model has legal prior knowledge and remains robust in subsequent reasoning, and introduces multiple modules during reasoning to integrate files, audio, and text. At the same time, it supports diversified output such as legal aid, legal documents, and mind maps.
At present, the team has open sourced three models, ChatLaw-13B, ChatLaw-33B and ChatLaw-TextVec, and continues to provide a free and available ChatLaw web version, users can use natural words to describe their encounters, and talk step by step Get legal help.
B station internal test "search AI assistant"
Enter "?" to experience
On July 3, Bilibili launched the internal test of the "Search AI Assistant" function. When users search for questions in the search box or enter "?" in the search term, they can experience this function.
This feature will provide an AI-generated answer to the user's search, and will attach a relevant reference video to the answer.
Chen Rui, chairman and CEO of Bilibili, said in a March earnings conference call that AIGC will greatly improve the search experience of Bilibili. Taking the numerous evaluation videos on station B as an example, users can get the most valuable content in the videos more quickly by asking AIGC.