Babbitt | Metaverse Daily Must-Read: Zhu Xiaohu and Fu Shengji debate GPT, what deep-seated contradictions are hidden under the surface of the AI boom? Is the large model "bubble" really about to "burst"?

Abstract: Under the upsurge of ChatGPT, Fu Sheng, chairman and CEO of Cheetah Mobile, had a debate with Zhu Xiaohu, a well-known investor and managing director of GSR Venture Capital Fund, whether entrepreneurship based on large models is valuable. This fierce debate has made more people see some deep-seated contradictions under the surface of the AI boom. In particular, the current growth rate of ChatGPT users is slowing down, the application of large models is not as good as expected, and domestic GPU chips cannot be filled and other factors are superimposed.

Image source: Generated by Unbounded AI tool

Hot info:

Beijing starts the work of sorting out the innovation map of the large-scale technology industry

According to the Securities Times e company report, on June 27, the Beijing Science and Technology Commission issued the “Notice on Soliciting Large-scale Model Application Cases and Application Scenario Requirements in the Artificial Intelligence Industry in Beijing.” The notice stated that in order to accelerate the high-quality development of social and economic development enabled by artificial intelligence technology represented by large models, we are now looking to Beijing’s artificial intelligence innovation entities and industry users to collect large-scale model application cases and industry application scenarios in the artificial intelligence industry. Initiate the combing work of Beijing’s large-scale model technology industry innovation map. Beijing will fully combine the needs of high-quality social and economic development and regional characteristics, focus on the innovation and application of large-scale model technology, accelerate the maturation and implementation of related technologies and products, continuously expand the breadth and depth of industry large-scale model applications, and form technical support, scene traction, and in-depth Empowering artificial intelligence industry large-scale industrial ecology.

OpenAI plans to make ChatGPT a “super intelligent personal work assistant” to compete with Microsoft

According to The Information, OpenAI CEO Sam Altman told investors privately that he plans to make ChatGPT a “super-intelligent personal work assistant” that can draft emails or documents for users based on their personal style and the latest business information.

But the assistant could create conflicts of interest with customers like Microsoft and Salesforce, which is also a major OpenAI partner, investor and cloud service provider, who also want to use OpenAI’s software to build artificial intelligence “copilots” (Copilots) to Auxiliary work. But building the new ChatGPT feature will be the focus of OpenAI’s commercialization efforts, according to Altman and two other people familiar with the matter. It is reported that in January this year, Microsoft announced that it would invest $10 billion in OpenAI and acquire a 49% stake in the company.

Jingdong will launch Yanxi AI large model, with parameters reaching hundreds of billions

According to the “Kechuangban Daily” report on June 27, at the JD Cloud City Conference Shanghai Station, Cao Peng, chairman of the JD Group Technical Committee and president of the JD Cloud Division, introduced that the upcoming large-scale pre-trained language model of Yanxi is a A new generation model with hundreds of billions of parameters. JD.com’s big model will be oriented to multi-modality and go deep into industrial scenarios such as retail, logistics, and industry. Cao Peng emphasized that in the era of industrial intelligence, a new generation of digital infrastructure is also needed, which can widely support various chip architectures and is compatible with various heterogeneous infrastructures.

Baidu Wenxin Large Model 3.5 already supports plug-ins and will be gradually opened to developers

According to a report by China Science Daily on June 27, Wang Haifeng, Chief Technology Officer of Baidu, said that the biggest change in Wenxin Model 3.5 is the addition of a plug-in mechanism. The default built-in plug-in “Baidu Search” enables Wenxin Yiyan to be generated in real time Ability to get accurate information; long text summaries and Q&A plug-in “ChatFile” supports super long text input and more. Wang Haifeng said that in the next stage, Wenxin Yiyan will release more high-quality Baidu official and third-party plug-ins, so that users can better use Wenxin large models. At the same time, Wenxin Yiyan will gradually open up the plug-in ecology to third-party developers to help developers create their own applications based on Wenxin’s large model.

China Mobile is building an “artificial intelligence platform” for general intelligence

According to a report from the “Kechuangban Daily” on June 27, at the China Mobile Technology Innovation Forum held today, Huang Yuhong, president of the China Mobile Research Institute, revealed that China Mobile is building an “artificial intelligence platform” for general intelligence. It is a new type of artificial intelligence infrastructure, carrying general intelligent large models and industry large models for national economic entities, based on ubiquitous computing network infrastructure, to achieve high-quality data utilization.

KPMG: Generative AI could boost UK productivity by £31bn a year

A new KPMG report has found that generative artificial intelligence has the potential to boost UK productivity by 1.2%, equivalent to an extra £31 billion in additional output to the UK economy each year, City AM reported on June 27.

Yael Selfin, chief economist at KPMG in the UK, says generative AI can “perform, on average, about 2.5% of tasks across UK occupations”, meaning workers will have more time to focus on other activities. The report also found that four in 10 jobs are expected to be impacted by generative AI. However, only 10% of occupations will face “significant impacts”. Generative AI won’t take away “many” jobs, but it could pose some transitional challenges, Selfin said. The KPMG report acknowledges that regulators and policymakers need to exercise “a high level of caution” because of the potential for “highly uncertain” wider social and economic impacts of generative AI due to its economic importance.

Radio Television Hong Kong launches AI weather anchor Aida, whose facial expressions and actions are synthesized by computer

According to IT Home news, on June 26, Aida, the AI weather anchor of Radio Television Hong Kong, officially debuted, reporting local and global weather conditions in the radio station’s “Radio Hong Kong Weather Briefing”. Radio Television Hong Kong stated on its official Facebook page that the AI anchor “will never eat screws (i.e. stuck, forget words, slurred speech)”.

According to relevant sources of the station, Aida is one of the experimental projects of AI applications in Taiwan. It took about 3 months to prepare and was produced by AI virtual avatar technology. In addition, Aida’s facial expressions, voice, mouth shape and movements are all completely synthesized by computer without the intervention of real people.

The largest cloud-based scientific research and intelligent computing platform in Chinese universities is online, supporting the accelerated operation of large models with hundreds of billions of parameters

According to a Shanghai Securities News report on June 27, CFFF, the largest scientific research and intelligent computing platform for domestic universities, was officially launched at Fudan University. This new type of “big scientific device” can realize efficient ultra-kilocal parallel intelligent computing and support the accelerated operation of large models with hundreds of billions of parameters, which is the first case among domestic universities. At the meeting, Fudan University signed strategic cooperation agreements with Alibaba Group and China Telecom respectively. Fudan University, together with China Telecom and Alibaba Cloud, will build a high-performance dedicated intelligent computing platform with the highest computing power among universities in the country. At the same time, Fudan University is about to hold the world’s first comprehensive scientific intelligence competition.

The World Economic Forum released the top ten emerging technologies in 2023, and generative artificial intelligence was selected

According to the Financial Associated Press, on June 26, the World Economic Forum released the “Top Ten Emerging Technologies in 2023 Report”, which selected the top ten technologies with the most potential and positive impact on the world. It is understood that these emerging technologies mainly include: flexible batteries, generative artificial intelligence, sustainable aviation fuel, wearable plant sensors, artificial intelligence-assisted medical care, etc. In the selection process of the top ten emerging technologies, a team of experts assessed the impact of technologies on people, planet, prosperity, industry and social equity. Experts believe these technologies, in addition to their significant benefits to society and the economy, are disruptive, attractive to investors and researchers, and expected to be deployed at scale within five years.

Investment and Financing:

Thomson Reuters acquires AI legal tech startup Casetext for $650 million

Thomson Reuters has reached a definitive agreement to acquire Y Combinator-backed AI legal technology startup Casetext for $650 million in cash, TechCrunch reported on June 27. It is reported that the transaction is expected to be completed in the second half of 2023, subject to regulatory approval and compliance with customary closing conditions.

It is reported that Casetext was founded in 2013, and its flagship product is CoCounsel, which aims to use AI to review documents, help process legal research memos, prepare forensics and analyze contracts. Casetext is one of a handful of companies granted early access to the GPT-4 language model, the infrastructure backend for CoCounsel. Casetext, which has 104 employees, has a client base of more than 10,000 law firms and corporate legal departments, according to a press release. Prior to the acquisition, the company had raised more than $64 million from Union Square Ventures and others. Thomson Reuters recently announced that it plans to spend about $100 million a year on artificial intelligence, incorporate generative artificial intelligence into its products in the second half of this year, and set aside $10 billion for mergers and acquisitions between now and 2025.

Metaverse recommended reading:

“Giant head-mounted display war “part ways”: Meta VR, Apple MR, Microsoft dislikes, Google gives up”

Compared with the hot and crowded AI concept, the head-mounted display track is particularly deserted, and Google terminated the company’s AR project. On Tuesday, Business Insider reported that Google terminated its augmented reality (AR) headset project, code-named Iris, and turned to building an AR software platform. At present, the head-mounted display giants have begun to “part ways”, Microsoft complained that the VR and AR markets are currently too small, and the XR business department has been disbanded at the beginning of the year. However, Apple and Meta are still insisting on MR. Earlier this month, Apple released the first head-mounted display “Vision Pro”, officially entering the MR track; Meta, which was once All in the Metaverse, continues to invest heavily, but has shifted its focus from the Metaverse to AI. .

“The latest AI rules of the U.S. Congress: No other products except the paid version of ChatGPT shall be used”

On Monday local time, some media obtained a memo signed by Catherine Spindow, the administrator of the US House of Representatives, explaining the latest requirements for the use of large language model applications within the US legislature. In the memo sent to all House staffers, Spindow explicitly requested that only the paid subscription version of ChatGPT Plus be allowed in House offices because the paid service includes “important privacy features that protect Congressional data.”

"Zhu Xiaohu and Fu Shengji Discuss Behind GPT, Is the AI Large Model “Bubble” About to Burst? "

The heated debate between Fu Sheng and Zhu Xiaohu in the circle of friends with different “thresholds” has made more people see some deep-seated contradictions under the surface of the AI boom. In particular, the current growth rate of ChatGPT users is slowing down, the application of large models is not as good as expected, and domestic GPU chips cannot be filled and other factors are superimposed. This round of domestic AI big model “bubble” really seems to be about to “burst”?

"Crush GPT-4! Google DeepMind CEO reveals: the next generation of large models will be integrated with AlphaGo》

Is the legendary Gemini, which merges AlphaGo and GPT-4-like large models, finally coming? One is the AI system that used reinforcement learning to defeat the human Go champion and created history. The other is the most powerful multi-modal large model that dominates almost all large-scale model lists. The combination of the two AIs is almost invincible up! Google DeepMind CEO Hassabis recently said to foreign media Wired that Gemini is still under development and it will take a few months, while Google DeepMind is ready to spend tens of millions of dollars, or even hundreds of millions. Previously, Sam Altman revealed that the cost of creating GPT-4 exceeded $100 million. Google DeepMind, of course, cannot lose.