AI Large Models Continue to Flourish Everywhere
-
After the popularity of ChatGPT, the development and launch of large models domestically and internationally have entered the fast lane.
Previously, Bianews published an article titled 'April: Domestic Large Models Flourish Everywhere,' focusing on the large model projects launched in March and April this year.
The boom continues, with multiple large model products emerging within a month. Beyond general-purpose models, some are more specialized, targeting industries, finance, education, transportation, and more.
In the 'iPhone moment' of AI, at the singularity point transitioning from the Mobile era to the LLM era, amid the dazzling frenzy of 'launch events,' questions remain unresolved: Are we ready for this AI arms race? Do we need so many large models?
On April 9, 360 officially announced that its AI product matrix '360 Zhiniao,' developed based on the 360GPT large model, has been deployed in search scenarios and will be open for beta testing to enterprise users.
It is reported that 360 Zhiniao is a search engine product based on AI technology. It employs advanced natural language processing to intelligently recognize user needs through voice and text input, delivering more accurate results. Additionally, 360 Zhiniao will deeply integrate with applications like browsers, smart marketing, Soda Office, and digital assistants to enhance user experience and productivity.
On April 21, Zhou Hongyi issued an internal letter requiring every employee, product, and business at 360 to fully embrace AI, adapt to human-machine collaboration, and begin reshaping products.
On April 26, Zhou Hongyi publicly demonstrated the iterative progress of '360 Zhiniao.' During the demo, he tasked 360 Zhiniao with writing a letter titled 'You Are So Useless,' imitating a parent addressing a child who doesn't love studying.
On the evening of May 7th, during a live discussion with Yu Minhong, Zhou Hongyi demonstrated the multimodal capabilities of the search engine-integrated AI product '360 AI Brain' in scenarios such as Q&A, writing, and text-to-image generation. Zhou stated that domestic large models would be 'boasting' if they claimed to surpass others without two years of imitation.
On February 10th, JD Cloud announced the launch of its industrial ChatGPT, ChatJD. The ChatJD intelligent human-computer dialogue platform is expected to have parameters reaching hundreds of billions. JD Cloud also unveiled the '125' roadmap for ChatJD's application.
The '125' plan includes one platform, two domains, and five applications. The platform refers to the ChatJD intelligent human-computer dialogue platform, a natural language processing platform for understanding and generating tasks, with parameters expected to reach hundreds of billions. The two domains are retail and finance, while the five applications include content generation, human-computer dialogue, user intent understanding, information extraction, and sentiment classification, covering the most reusable scenarios in retail and finance.
He Xiaodong, Vice President of JD Group, responded to the layout, stating that JD has rich scenarios and high-quality data in the ChatGPT field, such as JD Cloud's Yanxi interacting with users 10 million times daily.
Recently, Xueersi announced the development of its self-developed math large model, MathGPT, targeting global math enthusiasts and research institutions. MathGPT has already achieved phased results, and product-level applications based on this model will be launched within the year.
MathGPT aims to address three challenges of LLMs: solving problems correctly, providing stable and clear problem-solving steps, and making explanations interesting and personalized.
Xueersi has designated MathGPT as a core company project, led by CTO Tian Mi, and has established an overseas algorithm and engineering team in Silicon Valley.
On May 6, Taoyun Technology announced the launch of the Alpha Egg Children's Cognitive Large Model, which provides children with new interactive experiences in expression practice, emotional intelligence development, creativity stimulation, and learning assistance.
Liu Qingsheng, founder of Taoyun Technology, explained that as iFlytek's Spark Cognitive Large Model enters the R&D phase, Taoyun has incorporated its long-accumulated children's original language materials into the model. These materials cover children's stories, encyclopedic knowledge, and popular science readings.
Taoyun Technology has introduced active dialogue features, allowing robots to initiate topics related to children's experiences. Through multi-turn conversations, the system guides children to express themselves more confidently and frequently.
On April 20, Mobvoi announced the internal testing of its exploration large model "Sequence Monkey" at the 2023 AIGC Strategy Conference.
Mobvoi's "Sequence Monkey" is a large language model with multimodal generation capabilities. Its language-based ability system covers six dimensions: knowledge, dialogue, mathematics, logic, reasoning, and planning. It can simultaneously support various tasks including text generation, image generation, 3D content generation, voice generation, and speech recognition.
In addition to the large model, Mobvoi also launched a CoPilot product matrix for creators and an upgraded version of its consumer-facing voice assistant Magic Ask. The creator-oriented CoPilot product matrix includes an AI writing platform "WonderPen," an AI painting platform "WordsPaint," a voice dubbing platform "Magic Sound Studio," and a digital human video and live streaming platform "WonderMeta."
Li Zhifei, founder of Mobvoi, stated that their 'Sequence Monkey' AI has demonstrated emergent capabilities during training and is currently in an 'epiphany phase,' with future progress expected to accelerate.
Tesla CEO Elon Musk revealed in an April 17 interview his plan to develop an AI called 'TruthGPT,' designed to maximize truth-seeking and understand the nature of the universe. Musk noted the naming convention resembles Trump's Truth Social platform.
Musk has established X.AI, a Nevada-registered company where he serves as sole director, with family office manager Jared Birchall as secretary. The company is authorized to sell 100 million shares.
As part of his AI ambitions, Musk has been recruiting researchers to build an OpenAI competitor. Meanwhile, Chinese social platform Xiaohongshu reportedly formed its own large model team in March, led by Zhang Debing from their NLP advertising team, operating under strict confidentiality within the company.
Before taking charge of Xiaohongshu's large model project, Zhang Debing served as the head of intelligent multimedia algorithms at Xiaohongshu for a year, primarily focusing on AI and audio-video algorithms. Earlier, he was the leader of the multimodal intelligent creation team at Kuaishou, responsible for visual algorithm development.
In addition to establishing a large model team, Xiaohongshu has multiple independent departments advancing AIGC applications. In April, Xiaohongshu launched an AI creation app called "Trik," specializing in AI-generated art.
On May 6, iFlytek officially introduced the "Spark Cognitive Large Model," which boasts capabilities in seven dimensions, including text generation, language understanding, and knowledge Q&A. The model supports multi-style, multi-task long-text generation and can adapt writing styles to different contexts.
According to Liu Qingfeng, Chairman of iFlytek, the Spark model will be officially available to customers on August 15. By the iFlytek Global Developer Conference on October 24, the company aims for Spark to fully rival ChatGPT.
On May 8, iFlytek's stock surged to a limit-up, closing at 63.86 yuan. Its stock price has repeatedly hit record highs this year, with a year-to-date increase of 94.52%, nearly doubling, and its market capitalization reaching 150 billion yuan.
However, some netizens questioned whether the Spark model was merely a shell of OpenAI's ChatGPT, citing instances where the model claimed to be developed by OpenAI in conversations.
On May 11, iFlytek responded, stating that the Spark Cognitive Large Model was independently developed using vast training data. Due to ChatGPT's popularity, terms like "OpenAI" and "ChatGPT" frequently appeared in training data, leading to occasional incorrect responses. The company denied the claims as unfounded and illogical, arguing that if Spark were a shell of ChatGPT, it wouldn't outperform ChatGPT in response speed, text generation, or knowledge Q&A.
On May 5th, NetEase Youdao released a teaser video of its AI-powered spoken English teacher developed using the 'Ziyue' model. 'Ziyue' is a self-developed ChatGPT-like model designed for educational scenarios, aiming to provide students with more personalized and efficient spoken language learning services.
The video demonstrates that the AI teacher can offer various practice scenarios and assume multiple roles based on user needs, guiding users through multi-turn conversations to address the long-standing challenge of 'speaking anxiety' among Chinese learners.
NetEase Youdao stated in the comments section, 'The product is still in the development phase, and we will continue refining it to launch it at an appropriate time.'
Meanwhile, Alphabet plans to announce a series of generative AI updates at the Google I/O developer conference on May 10th, including the launch of a general-purpose large language model (LLM).
According to internal documents seen by the media, Google will introduce its latest and most advanced large language model, PaLM 2. This model supports over 100 languages and can perform a wide range of coding and math tests, as well as creative writing and analysis tasks.
Regarding large models, internal documents reveal that Google has been developing a multimodal version called 'Multi-Bard,' which uses a larger dataset to solve complex mathematical and coding problems.
Reports indicate that AI company Cloudwalk Technology will officially launch its large model product on May 18. The product will primarily be applied in smart finance, smart transportation, and other sectors mentioned in the company's previously disclosed private placement plan. Cloudwalk's large model product will cater to three directions: government, enterprises, and consumers, covering multiple fields such as finance, gaming, quality control, and transportation.
On the evening of March 30, Cloudwalk disclosed a private placement plan, proposing to issue up to 222 million shares to no more than 35 subscribers, aiming to raise no more than 3.635 billion yuan, all of which will be used for the development of Cloudwalk's "Industry Genie" large model project.
Since February this year, Cloudwalk has issued three stock price fluctuation announcements. It is reported that from the beginning of the year to the close on May 9, Cloudwalk's stock price surged by 151.47%, reaching its peak since listing in early April, with intraday prices nearing 60 yuan.
On April 18, mobile internet company APUS officially released its self-developed multimodal AI large model "AiLMe." According to reports, AiLMe has reached a scale of hundreds of billions of parameters, capable of understanding and generating text, images, videos, and audio.
For specific application scenarios, APUS has distilled four vertical models from AiLMe: the text model "Yique 8," the image model "Yique 3," the video model "Yique 4," and the audio model "Yique 6." Based on these, the company has innovatively developed a series of AI products such as "Smart Q&A Master, Sketch to Art, and Ink Wash."
It is reported that AiLMe will open API interfaces and additional services to customers. At that time, clients can call upon AiLMe's various AI technology capabilities based on their practical application needs.
On May 11, SoftBank Corp., the telecommunications subsidiary of SoftBank Group, announced its entry into the global competition to develop a ChatGPT version.
SoftBank CEO Junichi Miyakawa stated during an earnings briefing that the company established a new entity in March, selecting approximately 1,000 people to develop a Japanese version of OpenAI's AI chatbot ChatGPT.
Miyakawa also emphasized that SoftBank Group founder Masayoshi Son has long viewed artificial intelligence as a revolutionary force in how humans utilize technology. He further revealed that Son recently gathered a group of engineers to discuss the possibilities of ChatGPT.
Following this news, shares of several AI-related Japanese companies surged.