Sensetime Group, a leading Chinese company AI, waves in the world of technology with the announcement of it The latest AI model – Sensenova 5.0. The company's shares increased by more than 30% after messages, emphasizing the expectations and emotions associated with this breakthrough.
Let's look deeper at the key achievements of Sensenova 5.0:
- Knowledge improvement: Sensenova has undergone wide training, covering over 10tb of token data. This comprehensive training significantly improved its knowledge base, enabling better understanding and processing of complex information. The model adopts a mixture of expert architecture, enabling effective covering of context windows of about 200,000 when applying.
- Mathematics and reasoning: The mathematical possibilities and modeling of the model have been significantly increased. Now it can more effectively support numerical inference, code generation and a long understanding of the text.
- Language and creative skills: Creative writing, reasoning and Sensenov's abilities have improved significantly. Given the same knowledge, it provides better understanding, summary and the possibilities of questions and answers.
- Multimodal interactions: Sensenova supports parsing and understanding of high resolution images, as well as generating text for the image. It can distinguish complex data between documents and summarize the answers to questions, showing strong possibilities of multimodal interactions. His world -leading graphic and textual perception is first based on total results in authoritative references of multimodality.
The chairman of XU LI claimed that Sensenova 5.0 surpasses the GPT-4 OPENAI in most general use scenarios. Especially in Chinese Sensenova shines. His ability to understand, summarize and answer questions exceeds GPT-4 ability. This makes it an attractive choice for corporate applications, education and content generation.
In addition to Sensenova 5.0, Sensetime introduced the “cloud-to-crash” full-lane matrix in the industry. This matrix covers a large model on the cleaning side of the edges for terminal devices and integrated device of a large integrated model (enterprise). These models can be used in all areas, such as finance, coding, healthcare and government services, additionally expanding the horizons of the LLMS application.
