[December 22 Big Model Daily Collection] Emu2: a new generation of generative multimodal models; Midjourney can generate text! Five major upgrades of V6 version that amaze netizens

Nature | Discover new antibacterial compounds from 107 million molecules, MIT team develops DL method for antibiotic discovery

Link: https://news.miracleplus.com/share_link/14075
Currently, there is an urgent need to discover new structural classes of antibiotics to address the ongoing antibiotic resistance crisis. Deep learning methods help explore chemical space; these often use black-box models and do not provide chemical insights. Researchers at the Massachusetts Institute of Technology (MIT) have developed a deep learning method for antibiotic discovery and showed that it can identify potential antibiotics from large chemical libraries. The researchers used the method to discover halicin and abaucin from the Drug Repurposing Center, which contains about 6,000 molecules, and new antimicrobial compounds from the ZINC15 library, which contains about 107 million molecules.

Emu2: a new generation of generative multimodal models

Link: https://news.miracleplus.com/share_link/14076
It is a basic human ability to complete multi-modal tasks such as listening, speaking, reading, writing, and drawing based on a small number of demonstrations and simple instructions. For AI systems, how to utilize various information in a multi-modal environment to achieve multi-modal understanding and generation with few samples is a “technical high ground” that needs to be overcome. On December 21, 2023, Zhiyuan Research Institute released the new generation multi-modal basic model Emu2 as open source. Through large-scale autoregressive generative multi-modal pre-training, it significantly promotes breakthroughs in multi-modal context learning capabilities. Emu2 has significantly surpassed mainstream multi-modal pre-trained large models such as Flamingo-80B and IDEFICS-80B in the task of low-sample multi-modal understanding, and has performed well in many low-sample understanding tasks including VQAv2, OKVQA, MSVD, MM-Vet, and TouchStone. , visual question answering, agent-driven image generation and other tasks to achieve optimal performance.

Midjourney can generate text! Five major upgrades of V6 version that amaze netizens

Link: https://news.miracleplus.com/share_link/14077
Midjourney’s major update, V6 version is released! The biggest highlight after the update is that the images are more realistic, the details are processed more delicately, and it is more intuitive compared with the previous generation.

Microsoft Copilot will launch pinned chat conversation feature in January next year: up to 15 pinned chat conversations

Link: https://news.miracleplus.com/share_link/14078
Microsoft Copilot users will soon get graphics-based chat conversations that will make it easier for users to return to recent chats and better manage existing work. With the pinned chat conversation feature, users can pin up to 15 chat conversation windows to the top.

Yidu Technology releases self-developed medical model and signs cooperation agreement with Huawei

Link: https://news.miracleplus.com/share_link/14079
The “2023 Medical Intelligence Conference” hosted by Yidu Technology and co-organized by Huawei was held in Beijing. At the meeting, the large-scale medical vertical model independently developed by Yidu Technology was officially released, and Yidu Technology’s medical scenario solutions were comprehensively upgraded. In addition, Yidu Technology and Huawei signed a deepening cooperation agreement at the meeting and jointly launched smart medical solutions to accelerate the intelligent transformation of the medical and health industry.

Leyard Virtual Motion Point releases LYDIA professional ability action model

Link: https://news.miracleplus.com/share_link/14080
Leyard Virtual Motion Point officially released the “LYDIA” action model on the AI achievement release day. Huawei Cloud, JD Technology, iFlytek, Tianyi Cloud, Intel, Amazon Cloud and other partners attended the press conference to witness the historical moment of the AI era. Liu Yaodong, CMO of Leyard Group and CEO of Virtual Motion Point, released the LYDIA action model on-site. This model is a professional capability model independently developed by Leyard Virtual Motion Point. It has the “cognition, perception, and… Prediction, generation” and other core capabilities, it has taken the lead in exploring the new model of AIGC in the fields of spatial computing and action generation on a global scale.

Youzan White Crow: AI makes SaaS truly “available to everyone”

Link: https://news.miracleplus.com/share_link/14081
“In the past 11 years, we have spent a total of more than 3 billion on product design and R&D for these 7 systems.” In the past 11 years, Youzan has built a business operation system that includes 7 major systems and 20,000 functions. , the service scope covers social e-commerce operations, membership management, offline store management and other full-link SaaS systems. However, as systems become more complex and powerful, merchants often fail to make good use of these functions. Youzan founder and CEO Bai Ya believes that this is a problem faced by all SaaS companies, and it is also an unsolvable paradox for SaaS companies to B.