Recently, the popularity of ChatGPT almost overwhelmed the entire Internet. This phenomenon-level application with more than one million users in just 5 days after its launch (plus less than three months of testing), has been overwhelmed by the attention and popularity. When Alpha Go defeated the world Go champion, it was bound to leave a strong mark on the history of artificial intelligence.
However, Google, which has been making great achievements in artificial intelligence, can’t sit still. A few days ago, Google announced that it will launch a conversational artificial intelligence service called Bard, and also announced that the Google search engine will add more new artificial intelligence functions and services. This is seen as Google’s efforts to change the disadvantaged situation.
ChatGPT provokes nerves in the industry
According to SimilarWeb data, ChatGPT, which became famous overnight, has seen a rapid increase in the number of visits to the OpenAI website, and has now ranked among the top 50 websites in the world. In January this year, the number of visits to the OpenAI website exceeded 672 million, an increase of 3572% from November.
The heart-beating Microsoft quickly launched the latest version of the Bing (Bing) search engine and Edge browser supported by ChatGPT. The new Bing search will answer questions with a lot of context in a manner similar to ChatGPT. Microsoft CEO Nadella said that “the search engine has ushered in a new era.” Microsoft claims that Bing is built on a new next-generation large-scale language model, which is more powerful than ChatGPT, and can help it use network knowledge to intelligently connect with OpenAI’s technology. A new chatbot in Bing can help users refine their queries, draft and translate emails, rewrite computer code and more.
You know, Microsoft has been investing in OpenAI, the company to which ChatGPT belongs. On January 24, Microsoft and OpenAI officially announced that they will expand their partnership with “many years and billions of dollars” of investment, using the computing power provided by Microsoft’s Azure cloud computing service to achieve AI breakthroughs faster . In the future, all Microsoft products will have the opportunity to integrate ChatGPT.
Google CEO Pichai must have also been deeply touched, writing in his blog: “Many human questions are subjective and have no correct answers. AI can provide help and comprehensive insights at these times… Distilling complex information and multiple perspectives into an easy-to-understand format so users can quickly understand the big picture and get more information from the web.”
So Google quickly launched Bard, apparently in response to the challenge of ChatGPT. According to the introduction, Bard will use the lightweight LaMDA model (LaMDA model is Language Model for Dialogue Applications, a language model for dialogue applications), which was first announced at the Google I/O conference in 2021 and iterated to the second generation last year. . It is based on the Transformer neural network architecture and uses up to 137 billion parameters for training. Its training data is dialogue content, not ordinary sentences and articles.
In 2022, Google AI engineer Blake Lemoine even publicly stated that he believes that LaMDA already has a sense of personality and has the intelligence equivalent to an eight-year-old child. However, this statement was considered insufficient evidence by Google and many practitioners.
Google said that when users use Bard, they will collect external feedback and combine it with internal testing to ensure that “Bard’s answers meet high standards in key aspects such as quality, security and rootability.” It is worth noting that these three points are the three core goals that Google researchers value most in the LaMDA model.
OpenAI has also been working hard on the optimization of ChatGPT. Since the launch of the product, it has continued to improve ChatGPT’s acceptance of questions, quality of answers, product functions and mathematical capabilities, and has continuously consolidated its status as the most popular chat robot nowadays. moat.
False success may not be Google’s problem
According to a number of foreign media reports, in a recent demonstration, a user asked “How to tell a 9-year-old child about the discovery of the James Webb Space Telescope”, Google AI chat robot Bard replied: “James Webb Space Telescope shot The first pictures of planets outside our solar system.” However, this answer is not accurate. The European Extremely Large Telescope captured an image of a galaxy beyond our solar system in 2004, according to NASA records.
After Google’s AI chatbot Bard was found to have answered wrongly, the stock price of Google’s parent company Alphabet fell by more than 7% on February 8, and its market value evaporated by about 100 billion US dollars. However, Google is not the worst.
In 2022, after the launch of the academic AI robot Galactica, Meta was suspended for generating too much false information including racist content and incorrect articles within three days. What are the benefits?” Galactica also seriously provided the answer.
Meta has also launched products similar to ChatGPT and Bard before. Yang Likun, Meta’s chief AI scientist and winner of the Turing Award, has publicly shared the AI chat robot BlenderBot launched by Meta. The third-generation BlenderBot uses the OPT-175B language model with 175 billion parameters. The problem is not the ability, but that when BlenderBot was launched, Meta made many restrictions, such as being unable to answer sensitive questions such as politics and religion, making AI chatbots become A safe but boring product. Therefore, Yang Likun said that in terms of underlying technology, ChatGPT is not particularly innovative. The reason why it is so popular is mainly because the product design and combination are very good. He believes that ChatGPT’s grasp of reality is very superficial.
Still, Meta isn’t giving up entirely, as Joel Pino, head of its foundational AI R&D department, said: “AI is moving so fast, it’s unexpected. We’ve been keeping an eye on this space and making sure we have an efficient review process. However. , we have to catch up with the heat to make the right decisions and release the AI models and products that are most suitable for our community.”
In addition to Bard, on this basis, Google will also open up the Generative Language API (Generative Language API), which is also driven by the LaMDA model, for use by third-party developers and enterprises, thereby promoting more innovations. The API service is expected to be opened next month, and Pichai also said that Google will open more AI model-driven APIs and develop supporting tools.
Summarize
In short, Google, which has always been a leader in the search field and artificial intelligence technology, will definitely fight to the end this time. Its CEO Pichai issued a statement saying that Bard will use the information on the Internet to provide users with the freshest and high-quality responses. Google emphasized that Bard will combine the world’s information knowledge with the capabilities and intelligence brought by Google’s large language model. This means that Bard is constantly training its models based on real-time web data.
However, Mike Cook, an artificial intelligence researcher who studies computational creativity at King’s College London, is sober: “The technology is amazing. It’s interesting, which is what new technology should be. But it’s developing so fast, So much so that your understanding can’t keep up with its update speed. I think it will take a while for the whole society to digest it.”
After all, even with such convenient and fast work and study tools, human wisdom is not only presented in text, human emotions, judgments, aesthetics and creativity are still things that artificial intelligence is still beyond the reach of.
