Finishing | Dongmei, Nuclear Cola
Source | AI Frontline
Google’s Jeff Dean says that big corporations face more “reputational risk” than smaller companies when they provide misinformation, so Google is naturally “more conservative than small startups.”
Since the end of November, the AI chatbot ChatGPT was officially released to the public and quickly caused a sensation on Twitter, and Google employees naturally cast curious eyes. They wanted to see where Google stood in terms of cutting-edge chatbots that could answer users’ complex questions. After all, Google’s core business is web search, and it has long billed itself as a pioneer in the field of AI. Google’s dialogue technology is called LaMDA, which stands for Language Model for Dialogue Applications.
Google employees worry that their AI is no match for ChatGPT
At a recent all-hands meeting, employees raised concerns about Google’s competitive advantage in AI, given ChatGPT’s widespread popularity. After all, ChatGPT, which has attracted the attention of the world this time, came from OpenAI, a San Francisco-based startup backed by Microsoft.
At the meeting, such a question was widely concerned by employees, “Since LaMDA has been under development, does this mean that Google has missed the opportunity to gain an advantage in the field of chatbots?”
Alphabet Group CEO Sundar Pichai and Google’s long-term head of AI Jeff Dean answered this question, saying that Google is actually fully capable of producing similar results. It’s just that once something goes wrong, corporate giants like Google will undoubtedly need to bear higher economic and reputation costs.
Google’s search engine is used by billions of users around the world, while ChatGPT just hit 1 million users in early December. Dean explained, “The models really do address people’s needs, but at the same time, it’s also important to realize that there are still some real-world problems in these models.”
Google spoke and did not immediately respond to a request for comment.
ChatGPT is very strong, but Google is also “not easy to mess with”
Morgan Stanley published a report on Monday examining whether ChatGPT poses a threat to Google. Its Alphabet chief analyst Brian Nowak wrote that the ChatGPT language model could take market share and “upend Google’s position as an entry point for Internet users,” which is clearly not good for Google’s business.
However, Nowak is still confident in Google’s leadership position, because it has never stopped improving search, and this creative attempt is enough to deal with the impact of any new technology and competitive achievements. Additionally, Google is “building similar natural language models, such as LaMDA,” and “we expect more releases from Google over time.”
Pichai said at the meeting that Google has “many” plans in the field of AI chat in 2023. “This is an area that we need to explore boldly and be cautious and responsible. We must find a balance between the two ends.”
As a technology giant with a market value of more than 1.2 trillion US dollars, every move of Google must be extremely cautious. Dean told employees that, as of now, its technology remains largely in-house, emphasizing that a large company like Google would certainly be “more conservative” in dealing with “reputational risk” than a smaller startup.
“Of course we hope to apply these results to real products, to more language models, instead of hiding it behind the scenes for a long time. We have been working hard in this area, but many things not only need to be done, but also must be guaranteed. Do it well.” He added, “As you can imagine, for an AI chat application like a search engine, its authenticity will be very important; for other application scenarios, the bias, toxicity and safety issues of the AI model must be considered.”
Dean said that this technology is not suitable for widespread promotion, and the models that have been released so far have their own problems.
In Dean’s view, AI “can make up information, and they do. If the AI models aren’t sure about something, they’re going to make it up, like the biggest eggs are laid by elephants or something.”
As for Googlers, they already have access to its internal chat tool. During the COVID-19 pandemic, “a lot of people are talking to the system at lunchtime, and the whole conversation can be quite engaging.”
Pichai also said that 2023 will be an “inflection point” in the development of AI in conversation and search.
“We’re going to make big strides and launch more new products.”
OpenAI CEO admits ChatGPT is risky
Since its release, ChatGPT has been in the limelight, whether it is asking it to write a rhyming poem, check code for bugs, or answer scientific questions. ． ． ． ． ． It is easy to understand, the content of its reply is random every time, and the overall professionalism and information volume are maintained, which has a strong reference value.
ChatGPT is a fine-tuned GPT-3.5 series model with as many as 175 billion model parameters and was trained early this year. Model training is inseparable from the support of big data. The public crawler data set mainly used by OpenAI has a human language data set with more than one trillion words. In terms of computing power, GPT-3.5 is trained on the Azure AI supercomputing infrastructure (a high-bandwidth cluster composed of V100GPU), and the total computing power consumes about 3640PF-days (that is, one quadrillion calculations per second, running 3640 integers). day).
With such powerful performance, ChatGPT has 1 million users in just 5 days after its release, while the previous GPT-3 took nearly 24 months to reach this user volume.
But everything has two sides. After ChatGPT quickly became popular in the global technology circle, OpenAI CEO Sam Altman also admitted that ChatGPT is not perfect, and it also has limitations.
In a tweet over the weekend, OpenAI CEO Sam Altman also acknowledged the limitations of ChatGPT, saying that users should not rely too much on the answers given by this AI model.
“As it stands, ChatGPT answers must not be relied upon for any serious work. It represents a rehearsal and exploration, and we have a long way to go in terms of robustness and authenticity.”
Is Google okay?
In addition to chat AI, employees also have concerns about Google’s search business.
Except for a specific period in the new crown epidemic, Google has experienced the slowest business growth since 2013. Search-related revenue rose just 4% from the previous year, a much slower growth rate than the company’s overall advertising business.
At the meeting, Pichai read the following question aloud, “We don’t feel the same way now about headlines like ‘Google Search is dying.’ Sundar, are you worried about Google’s search business? What’s the underlying cause, and how do we deal with it?”
Pichai responded: “I think this is a good question. Google is still making progress, but people always feel that Google’s achievements are taken for granted. Only by looking forward to winning unexpected victories in the future can people be surprised and appreciated.”
Prabhakar Raghavan, senior vice president in charge of knowledge and information at Google, also responded. In July of this year, Raghavan publicly stated that as more and more young users turn to visual platform search, TikTok and Instagram have begun to erode Google’s share in the search market.
Raghavan told employees, “There is no denying that we have to step up and confront and address these issues. User expectations are constantly changing and they are asking us for new things. What we have to do is really build capabilities to meet those needs.” .”
The industry assessment also mentions that Google still has at least a 90% share of the search market and is still under scrutiny from regulators. Recently, Google executives have also become more open-minded, willing to discuss allegations of market monopoly and suppression of competition faced by Google’s search business.
Reference link:
https://https://ift.tt/CWHPcxm
https://ift.tt/zfEw27Z
(Disclaimer: This article only represents the author’s point of view, not the position of Sina.com.)
This article is reproduced from: https://finance.sina.com.cn/tech/csj/2022-12-15/doc-imxwttiv9793893.shtml
This site is only for collection, and the copyright belongs to the original author.