[Big Model Daily on December 28] Research Byte multi-modal large model PixelLM: efficient pixel-level reasoning without relying on SAM Industry Domestic embodied humanoid robot conquers complex scenes: real-time perception planning, dynamic movement bids farewell to “blind walking”

Byte multi-modal large model PixelLM: efficient pixel-level reasoning without relying on SAM

https://news.miracleplus.com/share_link/14497

ByteDance’s intelligent creation team, together with researchers from Beijing Jiaotong University and University of Science and Technology Beijing, proposed PixelLM, the first large-scale efficient pixel-level inference model that does not rely on SAM. Compared with previous work, the advantages of PixelLM are:

– Able to skillfully handle any number of open domain targets and diverse complex reasoning segmentation tasks.

– Avoiding additional and costly segmentation models, improving efficiency and migration capabilities to different applications.

Furthermore, in order to support model training and evaluation in this research field, the research team built a data set MUSE for multi-objective reasoning segmentation scenarios based on the LVIS data set and GPT-4V. It contains more than 200,000 Question-answer pairs involving over 900,000 instance segmentation masks.

ScienceAI 2023 “AI + Materials & Chemistry” Special Topic Annual Review

https://news.miracleplus.com/share_link/14498

The traditional method of scientific exploration is the “Edison-style” trial and error method. The material research and development cycle is very long, consumes huge manpower and material resources, and is subject to contingency. The design and research and development of new materials often face challenges: it is difficult to quickly screen and design urgently needed new materials, and it is difficult to find efficient and low-cost synthetic formulas for the designed new materials. New materials with synthetic formulas will also face a long cycle of scale. explore. Taking my country as an example, the Ministry of Industry and Information Technology’s survey of more than 30 large-scale backbone enterprises shows that of the 130 key materials, 32% are completely blank in the country, 54% can be produced, but their performance stability is poor, and only about 14% can be completely self-sufficient. . 2023 Large Model Thousand Sails Race.

Domestic embodied humanoid robot conquers complex scenes: real-time perception planning, dynamic movement bids farewell to “blind walking”

https://news.miracleplus.com/share_link/14499

This is the first time that a domestic humanoid robot dynamically ascends stairs and descends slopes based on real-time terrain perception. Faced with this “first time”, the humanoid robot named CL-1 is behind a Shenzhen company: LimX Dynamics. Founded in 2022, it has recently officially announced nearly 200 million in early financing to complete the capital level certification; now, the company has released the dynamic test results of CL-1, which shows a wave of their plans for full-size humanoids. New advances in robot evolution.

OpenAI Microsoft is sued! The New York Times accuses AI of infringement and requires the destruction of infringing models and training data

https://news.miracleplus.com/share_link/14500

OpenAI is facing perhaps the most difficult AI copyright lawsuit in history. The plaintiff, the New York Times, submitted a pleading with 220,000 pages of attachments to the district court. One section lists as many as 100 irrefutable evidences, and the output content of ChatGPT is almost identical to the news content of the New York Times.

Baidu CTO Wang Haifeng: The number of Wenxinyiyan users has exceeded 100 million

https://news.miracleplus.com/share_link/14501

At the Baidu WAVE SUMMIT + Deep Learning Developer Conference 2023 this afternoon, Baidu Chief Technology Officer Wang Haifeng announced that the current cumulative number of Wen Xinyiyan users has exceeded 100 million, and the number of daily questions asked by users every day is increasing rapidly.

Exclusive interview with the founding CEO of the Allen Institute for Artificial Intelligence: AI costs drop by half every 18 months, and the multi-modal era is coming next year

https://news.miracleplus.com/share_link/14502

Dr. Oren Ichioni, founding CEO of the Allen Institute for Artificial Intelligence and current board member and consultant, told reporters that Moore’s Law in the chip era still applies in the AI era, and the cost of AI training and inference may be every 18 dropped by half per month.