[Collection of Big Model Daily on March 15th] Apple’s big model MM1 has entered the market: more than half of the authors are Chinese; setting a precedent for AI drug research, Insilico disclosed the entire process of research and development of its first AI drug, published in a Nature sub-magazine; Haier and Leju The humanoid robot “Kuafu” is on display: it supports operations such as washing clothes and watering flowers; was Sora trained using this data? OpenAI CTO’s confession arouses public outrage; OpenAI official document leaks, GPT-4.5 turbo may be released in July

Apple’s large model MM1 is entering the market: 30 billion parameters, multi-modal, MoE architecture, more than half of the authors are Chinese

Link: https://news.miracleplus.com/share_link/21115

Since this year, Apple has obviously increased its emphasis and investment in generative artificial intelligence (GenAI). Previously at the 2024 Apple Shareholders Meeting, Apple CEO Tim Cook said that significant progress will be achieved in the field of GenAI this year. In addition, after Apple announced that it would abandon its 10-year-old car-making project, some members of the car-making team also began to turn to GenAI. In this way, Apple has conveyed its determination to add GenAI to the outside world. GenAI technology and products in the multi-modal field are currently very popular, especially represented by OpenAI’s Sora. Of course, Apple also wants to make achievements in this field. Today, in a paper “MM1: Methods, Analysis & Insights from Multimodal LLM Pre-training” signed by multiple authors, Apple officially announced the results of its own multimodal large model research – this is a model with up to 30B parameters The multimodal LLM series.

Improving generative zero-shot learning capabilities, the visually enhanced dynamic semantic prototyping method was selected for CVPR 2024

Link: https://news.miracleplus.com/share_link/21116

In existing generative zero-shot learning methods, when the generator is trained and used, it is conditioned on Gaussian noise and the semantic description of the entire category, which limits the generator to only optimize for the entire category. Rather than describing each sample instance, it is difficult to accurately reflect the distribution of visual features of real samples, resulting in poor generalization performance of the model. In addition, the visual information of the data set shared by seen classes and unseen classes, that is, domain knowledge, is not fully utilized in the training process of the generator, which limits the transfer of knowledge from seen classes to unseen classes. In order to solve these problems, graduate students from Huazhong University of Science and Technology and technical experts from Intime Business Group, a subsidiary of Alibaba, proposed a visually enhanced dynamic semantic prototyping method (called VADS) to more fully introduce the visual features of seen classes into semantic conditions. Promoting generators to learn accurate semantic-visual mapping, the research paper “Visual-Augmented Dynamic Semantic Prototype for Generative Zero-Shot Learning” has been accepted by CVPR 2024, the top international academic conference on computer vision.

In 18 months, setting a precedent for AI drug research, Insilico disclosed the entire process of research and development of its first AI drug and published it in the Nature sub-journal

Link: https://news.miracleplus.com/share_link/21117

In June last year, Insilico Medicine announced that the world’s first AI-designed drug, INS018_055, entered Phase II clinical trials and was administered to the first patients. Recently, Insilico published an article titled “A small-molecule TNIK inhibitor targets fibrosis in preclinical and clinical models” in “Nature Biotechnology”, focusing on the entire process of what it claims to be the first AI-generated and discovered drug INS018_055. “The progress of INS018_055 is a proof-of-concept for Pharma.AI, our end-to-end AI-driven drug discovery platform, and sets a precedent for the potential of generative AI to accelerate drug discovery,” said Alex Zhavoronkov, founder and CEO of Insilico.