9 large AI models were approved to go online
The "Hundred Models" war has officially begun
Spelling algorithms, fighting for data, computing power...
What should the big models compete with?
They will bring changes to life and industry
What surprises and challenges does it bring?
In August this year, 8 large model registrations were approved. Baidu, Byte, Zhipu Huazhang, SenseTime, Chinese Academy of Sciences, Baichuan Intelligence, MiniMAX, and Shanghai Artificial Intelligence Laboratory became the first batch of lucky ones in China and took the lead. Chance. The most direct benefit of filing is that it can be opened to a large number of C-side users, thereby having the opportunity to obtain more user feedback.
Today, the second batch of lucky winners in China were born, including big names from 9 companies including Dark Side of the Moon, Kunlun Wanwei, Zhihu, Meituan, Mobvoi, Face Wall Intelligence, Kingsoft, NetEase Youdao, and TAL The model is officially approved. Among them, the first 6 companies prefer general models, and the last 3 companies focus on vertical fields.
※The following list is in no particular order:
Up to now, almost all large models from major domestic manufacturers have been registered. It can be predicted that in the next 1 to 2 years, the general large model track will be rapidly saturated. We will wait and see who can become the “ChatGPT” in China.
Moonshot AI
As a strong competitor of the Chinese version of OpenAI, Moonshot AI founded by Yang Zhilin has achieved breakthroughs in the field of "long text". On October 9, the company officially launched Kimi Chat, the first smart assistant product that supports input of 200,000 Chinese characters. This is the largest model service currently available in the global market that can be used in product form. The maximum context input length.
Long text opens a new world of large model applications
How will a large model perform with very long context input? Here are some examples of Kimi Chat in action:
Leave the long article of the public account directly to Kimi Chat and let it help youQuick summary and analysis:
Leave the freshly released NVIDIA financial report to Kimi Chat for quick completionKey information analysis:
Too many business trip invoices? Drag them all into Kimi Chat, quickly organize them into the required information:
Through the above examples, we can see that when the context that the model can handle becomes longer,the ability of the large model can cover more usage scenarios, which really matters to people. play a role in work, life and study.
Education companies are rushing to join the game
The application of AI can make education more intelligent, help students learn in a personalized way, and improve the quality of teaching. Against this background,many education companies have intensively entered the AI large model track and launched related products based on their previous advantageous businesses.
01 Star Fire Knowledge Large Model
In May, iFlytek released the Spark Cognitive Model. Liu Qingfeng, chairman of iFlytek, introduced at the scene, " Large model + AI learning machine" allows AI to correct essays like a teacher and have real-life conversations like a speaking teacher. The upgrade of these two functions will promote language learning to a whole new level.
iFlytek's Spark cognitive model has reached the world's top level in terms of text generation, knowledge question and answer, and mathematical abilities.
Enter the knowledge points you want to learn, and it will formulate learning directions and ideas for you:
Leave it with obscure and convoluted logical reasoning questions, and it can alsoaccurately identify:
(A says "B is lying", B says "C is lying", C says "A and B are both lying", then who is telling the truth?)
Faced with high school physics questions, the Spark Cognitive Model alsoselected the correct answer and explained it:
02 Confucius said big model
In July, NetEase Youdao launched "Zi Yue", a large vertical model in the education field., and then in August it released " Three new hardware products are launched for the first time in the "Ziyue" educational model - Youdao Dictionary Pen X6 Pro, Youdao Dictionary Pen S6 and Youdao Hearing Treasure Pro.
(※ Ziyue Big Model is currently not open to internal testing qualifications)
03 MathGPT large model
In August, during TAL’s 20th anniversary live broadcast, CTO Tian Mi officially announced that the company’s self-developed MathGPT, a 100-billion-level model in the field of mathematics, was officially launched< a i=2> and open public beta.
Since it is a large model dedicated to mathematics, the dialogue content can currently only answer questions related to mathematics.
The question was recognized in the dialog box, and the result was not only accurate, but also provided specific formula steps and detailed analysis:
The iFlytek Spark model and MathGPT have demonstrated strong capabilities in multiple fields, especially in text generation, language understanding, and pan-domain knowledge question and answer. However, it should be noted that it still has some limitations, such as its performance on complex problems such as code generation and semantic reasoning needs to be further improved.
AI Big Model: Continuous Exploration
After a comprehensive review of the development of large AI models in 2023, it is not difficult to find that, whether in the fields of natural language processing, image recognition or multi-modal interaction, the AI products of major companies are constantly breaking through themselves and pushing the entire industry forward. develop.
Looking to the future, AI large models will continue to explore on the road to intelligence, bringing more possibilities to the world.