To The Moon

Faced with the pressing forces of Open AI and $Microsoft(MSFT)$ , $Alphabet(GOOG)$ Google finally stopped sitting still.

Today, Google announced the official opening of the public beta of its ChatGPT competitor Bard, which will first be launched for the United States and the United Kingdom. At present, users in these two countries can wait in line on bard.google.com. As the test progresses, Bard will gradually go online in other regions.

However, compared to the complacent state of Open AI and Microsoft, Google, which had previously "turned over" in Bard, was obviously more cautious. Different from New Bing's large-scale open strategy, Bard's test quota will be gradually released, and the initial version will only respond to text.

This time, Google also showed you more information about Bard's user interface and functions. Next, let's take a look at how we will use Bard on Google in the future, and how it is different from New Bing.

Same AI Algorithm, stronger "survival desire"

Overall, the usage patterns and functions of Google's Bard are basically the same as those of New Bing. This time, Bard also provides users with a separate chat interface. Users can ask questions and search in the text box, such as asking Bard to help you write a plan, list a recipe, and help you search for a cold knowledge.

But in terms of positioning, Google emphasizes that Bard is not a substitute for its search engine, but a "complement to search". Bard calls itself "your creative and capable collaborator (Collaborator)", while New Bing calls the AI assistant It's called the "Copilot".

When you ask a question on Bard, there are four buttons at the bottom of the answer, which are Like, Dislike, Regenerate and Google it. In other words, you can seamlessly connect to Google's main search interface in the chat interface, and the two will complement each other.

However, New Bing is currently able to provide information sources for AI-generated answers, and the function of automatically providing extended questions based on generated answers is currently not available in Bard. This may also affect the credibility of Bard's information output to a certain extent, and bring limitations to the expansion of search information scenarios.

New Bing's AI conversational bot interface

But Bard also does more than New Bing. When a user asks a question, Bard can directly help the user generate 3 different versions of the answer. You only need to select "View other drafts" to switch between these drafts and choose according to your content preferences.

Generally speaking, in terms of AI functions, Bard is similar to New Bing. However, based on the lessons learned from the chaos and criticism that occurred after the launch of New Bing, the biggest feature of Google Bard this time is that the entire interface reveals Google's full "desire to survive".

Whether it is under the text box where your Bard asks questions or under the answer generation box, there is always a line of small characters written to remind users that "Bard may generate inaccurate or slightly offensive information, but it does not represent Google's official opinion." Before you start using Bard, you will also be clearly reminded that Bard is still in the experimental stage. There is also a big experiment sign next to the Bard logo on the page.

In today's official blog, Google even gave an example of Bard's wrong answer to give users a "vaccination" and let everyone manage their expectations.

With Google's "hope of the whole village", will Bard be able to avenge its shame?

Google's caution this time is entirely understandable, after all, a factual error in Bard's debut a month ago cost Google $120 billion in market value. At that time, Google had vowed that Bard was not only competent for all the roles of ChatGPT, but also more responsible than ChatGPT, but it did not expect to overturn the car immediately in the demonstration. Not only Bard was ridiculed by the group, but Google has also been "reported" by Microsoft recently. Press and beat".

This time, Google has obviously learned the lessons of the last time when it opened the public beta Bard, and its attitude is more modest, and the information it releases is more comprehensive and sufficient. It is understood that after the Bard overturned, Google CEO Pichai issued an internal letter to all employees of the company. On the one hand, he told everyone to stabilize their mentality. guide.

In the guidelines, Google highlights how to build a "responsible AI" that lets employees ask Bard questions and evaluate their responses, flagging inappropriate responses if they don't match the facts. It also clarifies the do's and don'ts, including not describing Bard as a person, not making emotional insinuations, etc. It is particularly emphasized that when it is found that AI generates content that is hateful, harmful, false, illegal, or abusive, or that solicits sensitive information, it must be flagged and reported to the team.

Similar to the current open public beta interface, employees also have "like" and "dislike" buttons during the test period, which can help quickly classify information. According to news, in the past few weeks, 80,000 Google employees have participated in the test, and many employees have to spend 2-4 hours a day to talk to Bard and provide test feedback.

In an internal meeting, Pichai hinted that Google is not necessarily in a hurry to launch Bard, but must first ensure the reliability of the technology. Now that Bard has officially opened the public beta, it seems that Bard has passed some internal review standards in the past few weeks.

When asked about the difference between himself and GPT-4, Bard showed considerable confidence. Bard said that although he and GPT-4 are both large-scale language models, he used 500 billion parameters for training, but GPT- 4 is only 175 billion, so I can produce more complex and relevant content than GPT-4.

Although there is no factual error, the answer given by New Bing is biased and the logic is confusing. First, it said that the training data sets of Bard and ChatGPT are different in time, and then they said that experts believe that GPT-4 will let ChatGPT see It sounds like a joke, and the answer gradually moves away from the question being asked.

Judging from the performance of this issue, Bard's content logic structure is obviously better than that of New Bing, but from the perspective of information reliability, New Bing has at least made no mistakes so far.

ing's daily activity has exceeded 100 million, and Google is trying to catch up

As we all know, Google got up early and rushed late in the AI battle. The recent fierce offensive launched by Open AI and Microsoft can be said to have pushed Google to the edge of the cliff.

At the beginning of this month, when New Bing was launched for a month, Microsoft officially released the latest statistics, saying that New Bing has attracted more than one million new users in the past month. With this promotion, Bing’s daily active users now It has passed the 100 million mark. Among them, about one-third of users use Chat's functions every day, with an average of about 3 chats per session, and a total of more than 45 million chats.

In addition to search, in the browser field dominated by Google Chrome, Edge, which integrates AI capabilities, has also begun to rob users. Microsoft's "search + answer + chat + creation" ecosystem is gradually taking shape.

For Microsoft's attack, Google can't be in a hurry. During this period of time, the two companies are basically in a state of coming and going, seeing each other, as long as one holds an AI conference or announces a new product, the other will definitely release new news the next day. Now the flames of war between the two have burned all the way from search and browsers to the field of office software.

For Google, which is already one step behind, it is not easy to break through Microsoft's encirclement in a short time. But in the field of search, Google's unique geographical position also makes it "easy to defend and difficult to attack". As long as Bard does not frequently roll over or have any outrageous incidents, it will be difficult for Bing to pose a great threat to it.

At the same time, aiming at the current technical defects of ChatGPT and focusing on solving the problem of information security and reliability may also become an opportunity for Google to overtake in a corner.

Regarding the threat of ChatGPT, Pichai has previously stated that some of Google's most successful products in the past were not the first to be introduced to the market. The reason why it was able to succeed in the end is that Google paid attention to the needs of users, built on deep technical insights, and finally won the trust of users over time. In terms of AI competition, Google will remain the same.