First, let’s laugh at the fact that Google’s own AI tool, developed by Google, actually advocates spinning off Google to stop a monopoly? On the issue of anti-monopoly, Google’s generative AI Bard, which has just been opened for public testing, is quite righteous.
Bard actually supports splitting Google
The Bard public beta is long overdueOn Tuesday, Google announced the beta release of its generative AI Bard, which will be gradually opened to the registered public in batches. But for now, the beta is only open to users in the US and UK, and for now, only English is supported. Users in other regions and languages will need to wait for subsequent updates to the Bard.
Although Google has a long product line spanning many areas, Bard is undoubtedly the most talked about and important new product in recent years, and is directly related to Google’s competitiveness and market prospects in the coming years. Although Google publicly demonstrated Bard in early February, it was not open for public testing, but only to Google employees and a few invited and trusted beta testers.
During Google’s demonstration of Bard in early February, there was a factual error (a misrepresentation of the basics of the Webb astronomical telescope). Although this minor error was quite specialized and only astronomy enthusiasts could have caught it, it still dragged Google’s parent company stock down 8% and evaporated $100 billion in market value in one day.
Google CEO sends internal memo
Perhaps because of this previous experience, Google executives are very cautious about Bard’s public testing. After opening the public test, Google CEO Pichai issued an internal memo to employees on Tuesday saying, “We’ve taken a responsible approach to research and development, including inviting 10,000 trusted people from a variety of backgrounds and positions to test, and we welcome all feedback to learn from it and keep iterating and improving.”
It looks like Pichai is taking a precautionary shot at possible Bard slip-ups. “Even after all this progress, we’re still at the beginning of a long AI journey. As more and more people start using Bard and testing performance, they will surprise me with the problems that may arise.”
At the same time, Google is downplaying the impact of generative AI on its own search business, emphasizing that Bard is not a replacement for the search engine, but rather a complementary aid. It’s worth noting that Bard is a separate web dialog box, rather than serving as a Google search tool, but will offer a Google search button after the answer.
ChatGPT Applies Heavy Pressure
The hottest product in the tech industry in the past three months or so is undoubtedly ChatGPT, a generative AI product created by OpenAI that has been released worldwide, giving users an unprecedented smart experience and directly driving the next tech windfall, even to the Chinese Internet venture capital community.
On the other hand, the Internet giant Microsoft has been increasing its strategic investment in OpenAI, tying it firmly to its Azure cloud platform, and then connecting it to its office component and Bing search after ChatGPT became a big hit, making AI the core competitiveness of its own products and launching another attack on Google, which dominates the search market.
Behind this ChatGPT boom, there is no one more embarrassed than the industry giant Google. Although Google started developing generative AI as early as 2015 and released the conversational application language model LaMDA in 2021, it has been steadily developing it internally.
In June last year, a Google AI engineer claimed that LaMDA already had human emotions, and although Google quickly moved him out of his job, the matter was exposed by the media, which still sparked considerable controversy and speculation, and made Google even more cautious about AI security and unwilling to launch in haste.
After ChatGPT was released last year, AI project leader Dean (Jeff Dean) said in an internal meeting that although Google has similar AI tools, the technology is still immature and if the product is launched too hastily, it may affect Google’s reputation, as users trust the information provided by Google very much.
However, Google now has to speed up the pace of research and development to catch up with competitors’ products. This is not the pace Google is used to. Last December, Google raised the product level for generative AI to the highest red, coordinating its internal AI R&D resources to accelerate the launch of its own response products.
In fact, Google is in a quandary about the pace of Bard’s advancement. If the research and development action is slow, the outside world will worry about Google’s competitive strength and industry prospects, dragging down the stock price; but if the promotion is too hasty, the outside world will be more disappointed with Bard’s mistakes, depressing the stock price tumbling. This is perhaps the main reason for Pichai’s frequent precautions in Bard’s public testing.
The biggest advantage of real-time networking
So, how does Bard perform in practice?
Compared to ChatGPT, the biggest advantage of Bard is the huge information resource, they can rely on Google’s search tool to get the latest information and give answers based on their own combing. Previously, ChatGPT’s training data was only up to 2021 and could not answer information inquiries after 2022. In addition, Bard’s response time is significantly faster than ChatGPT’s.
Bard can search for information online, meaning that users can directly ask for today’s sports scores, stock price movements, and news series updates. In real-world testing, however, I found that Bard still has a time lag and is not yet fully real-time.
Still, Google Bard’s networking capabilities have a clear advantage over ChatGPT’s training database, and they can give their own answers on the latest events. When I asked about the possible recent indictment of former U.S. President Donald Trump in New York and the Silicon Valley bank collapse, Bard gave a brief but relatively complete rundown and summary of events.
Let Bard find restaurants
Helping users with flight and hotel reservations is a job for Google. But Bard can be misinformed. When I asked Bard on Tuesday to book myself a flight from San Jose, California to Charlotte, North Carolina with the shortest flight time, Bard gave me the wrong flight number. When I asked Bard again on Thursday, I found that Google had corrected the error and gave me the correct flight number.
ChatGPT supports third-party plugins for real-time information
Just today, however, OpenAI announced that it is allowing ChatGPT to access real-time information in a controlled environment through certified and secure third-party plugins, solving the biggest pain point of not being able to connect to the Internet before. The first batch of plugins are mainly from Expedia, Instacart, Shopity, Slack and other services to help users access life and collaboration information. This means that ChatGPT can also provide food and ticket ordering, as well as retrieve company documents and notes information.
The Great White Spin-Off Google
Bard writes poetry in the vernacular
Doing math, telling a joke, writing a love poem – these conversational actions are already a piece of cake for ChatGPT, and they’re no big deal for Bard. Bard also allows users to copy text.
Bard answers how to comfort a sick boy
In terms of work assistance, Bard can also perform daily tasks, helping users write business emails and simple articles, as well as make forms, take notes, and do analysis, but it requires uploading files to Google Drive.
Although Bard provides voice input on the web side, it does not work in practice; although Bard claims to support multiple language translations, actual use indicates that it cannot complete commands at the moment. All these details show that Bard is still a half-baked product and needs to be improved.
Ask Bard for his review of The Instant Universe
I don’t know programming, so I can’t experience the difference between ChatGPT and Bard in terms of programming ability. But from the feedback of my programmer friends around me, ChatGPT seems to understand their needs better in this regard, “Bard doesn’t give the syntax highlighting on black background that coders need. The quality of answers to Python questions is also far worse than ChatGPT.”
Bard does have its own opinion output. Big Language Models (LLMs) like Bard are trained on publicly available content, and they reflect positive or negative views of specific politicians, celebrities, and other public figures, and even give opinions on certain aspects of controversial social or political issues in their answers, a Google spokesperson said after the open public beta.
However, Bard many times does not display the source of the information, so accuracy is left to the user’s discretion. Put another way, it is not clear to the user how Bard reaches its conclusions, and there is a risk of being misled by Google’s algorithm into believing its concluding positions. In the context of a deeply divided American society, this is likely to raise questions from conservatives.
Interestingly, when one of the testers asked Bard what he thought of the U.S. Department of Justice’s antitrust lawsuit against Google, Bard took a strong stand and chose to be “righteous”. restrictions on its advertising transactions. These measures have harmed consumers and stifled innovation in the digital advertising marketplace. I hope the court will endorse the DOJ’s claim and order Google to take steps to break the monopoly.” It appears that Bard did not give its parent company special treatment just because Google produced it, and Google did not interfere with Bard’s views on the most sensitive antitrust issues.
AI Arms Race Escalates
Perhaps if Bard had been the generative AI released first late last year, Bard would have done well enough to spark a wave of admiration and perhaps an industry boom. Bard is not yet available in other languages, it does not support images, it is not programmable, and users in international markets do not know when it will be available.
Just a few days ago, OpenAI released GPT-4, a new generation of large multimodal model that accepts image and text input, provides text output, and is trained on more data (based on the Microsoft Azure platform) and has more weights in the model file. In the words of OpenAI, GPT-4 already performs at a human level on a variety of professional and academic benchmarks.
GPT-4 also surpasses many previous versions in creativity and collaboration, and surpasses ChatGPT in advanced reasoning. according to OpenAI, GPT-4 has been scheduled to take the U.S. Bar Exam, the Law School Admission Test, and the U.S. Advanced Placement SAT in math and writing. On these tests, GPT-4 outperformed 88% of test takers.
Not surprisingly, Microsoft was the first to launch a preview version of its Azure OpenAI GPT-4 service for the enterprise. Enterprises can build their own applications based on the Azure OpenAI service. Robot developers can build virtual assistants in natural language in just a few minutes.
In this wave of generative AI arms race, OpenAI is undoubtedly the temporary leader, one step ahead of Google in terms of both pace of development and intelligence; it also gives its strategic partner Microsoft a glimpse of catching up with Google. Since Bing was connected to ChatGPT in early February, the number of Bing page visits has increased by 15.8%, the number of global app downloads has increased by 8 times, and the number of daily users has exceeded 100 million. At the same time, Google search page visits are down 1% and search app downloads are down 2%.
The search market, which was already a foregone conclusion, has also been reinvigorated by generative AI technology. However, this does not mean that Google’s market position is about to be shaken. Google Search still has more than 1 billion daily users, more than 10 times that of Bing. If Google improves the user experience of Bard as soon as possible, opens a full public test to global users, and supports multiple languages, with its huge user base and data advantages, the future AI arms race will still be difficult to determine.
Finally, I asked Bard who his main competitors were, and Bard listed Microsoft’s Ice, OpenAI’s GPT-3, and Amazon’s Lex. “All of these models are trained on large amounts of text and code to generate text, translate language, write multiple creative content, and answer user questions. But I believe my unique ability to acquire and process information from the real world through Google search consistently keeps the responses consistent with the search results. Thus being able to answer your questions in a more comprehensive and informative way.”