The search engine operator, founded in 2000, now describes itself as a leading provider of artificial intelligence with a strong Internet business.
(Photo: REUTERS)
Beijing China’s largest search engine operator Baidu will launch an intelligent chatbot service called “Ernie Bot” in March. The company announced this on Tuesday. The ChatGPT-like offer is to be embedded in the main search services. The project, which is called Wenxin Yiyan in Chinese, is currently being tested internally.
Baidu wants to speed up the rollout to keep up with US search engine providers Google and Microsoft, the company said. On the Hong Kong stock exchange, Baidu shares rose more than 17 percent at times after the announcement.
Tech companies worldwide have high hopes for this form of artificial intelligence (AI). Microsoft intends to invest around ten billion dollars in the start-up Open AI in the coming years. Its chat program ChatGPT has triggered a real hype in the past few weeks. Google announced on Monday evening that it would make its chatbot “Bard” publicly available in the coming weeks.
Baidu chatbot was also trained with infographics
ChatGPT understands user questions in natural language and generates complex texts on command. It uses data from books, newspaper articles or social media. However, the answers of the so-called generative AI can also be incorrect.
In addition to texts, the Baidu chatbot is said to have been trained with data from infographics or other visual forms of representation as well as with Chinese and English-language data sets. He is part of the “Ernie” project, which is based on Baidu’s deep learning platform Paddle Paddle. Ernie stands for “Enhanced Representation through Knowledge Integration”, which can be translated as an improved (result) representation through the integration of knowledge.
>> Read also: How Open AI’s chatbot works
In the project, Baidu summarizes its generative AI offerings, for example in the areas of language understanding, speech generation and text-to-image generation. According to the company, a total of 36 of the models have already been released. One example is “Ernie-ViLG”, which allows Internet users to create artificial images by entering text. Open AI has a similar offering called DALL-E.
Baidu wants to be a leader in generative AI
Baidu CEO Robin Li internally in December cited ChatGPT as an example of a technology his company could become a leader in, Bloomberg quoted a transcript as saying. However, he also cautioned that commercializing the technology is a challenge. Baidu is said to have started development of the “Ernie Bot” in September.
The search engine operator, which was founded in 2000, describes itself as a leading provider of artificial intelligence. There is no other company that can compete with Baidu in natural language processing, they confidently emphasize. The group has invested billions in research into AI in recent years and is increasingly relying on future technologies such as quantum computers and autonomous driving.
>> Read also: A search engine giant is now building cars – With these models, Baidu is pushing onto the market in 2023
In China, researchers, companies and investors are pushing the development of content created with artificial intelligence. In the AI index of the US elite university Stanford, which evaluates global progress in the areas of research, development and economy, China comes in second, just behind the USA. In some areas, the People’s Republic is even ahead.
China’s internet censors unveil AI content bill
However, like traditional and social media, AI-generated content is also subject to censorship in China. A sample of the specialist publication “MIT Technology Review” – published by the US elite university Massachusetts Institute of Technology (MIT) – revealed that Baidu’s text-to-image offer “Ernie-ViLG” contained numerous politically sensitive keywords such as Tiananmen and Xi Jinping filter out.
In January, Internet regulator CAC introduced a new bill to ban AI content that “endangers national security and social stability.” In addition, the respective providers should be liable for the dissemination of misinformation. In addition, most digital offers can already only be used if you register with Wechat or a cell phone number and are thus identifiable. In addition to censorship from outside, this often leads to self-censorship by users.
ChatGPT can currently only be used indirectly in China. Registration with a Chinese mobile phone number is not possible. Anyone who manages to gain access will quickly notice that the quality of the answers in Chinese lags behind that in English. This is because ChatGPT was mainly fed with English-language training data. It shows the challenge for AI models trained on specific speech datasets.
More What Microsoft intends to do with ChatGPT.