[December 26 Big Model Daily Collection] Huawei Noah’s Pangu Agent is here, allowing agents to learn structured reasoning; Moonshot Al chatbot Kimi Chat has been updated, equipped with new models, supports scanned PDFs, etc.; Vincent video software Pika 1.0 Open web access to all users today

Huawei Noah’s Pangu Agent is here, allowing agents to learn structured reasoning

Link: https://news.miracleplus.com/share_link/14329

Recently, researchers from Huawei’s Noah’s Ark Laboratory, University College London (UCL), Oxford University and other institutions proposed the Pangu Agent framework (Pangu-Agent) to try to solve the challenges faced by AI agents. The study authors include Wang Jun, a professor at the Department of Computing at University College London.

Lower the rank of the Transformer and remove more than 90% of the components in a specific layer without losing LLM performance.

Link: https://news.miracleplus.com/share_link/14330

Researchers from MIT and Microsoft made a surprising discovery in the paper “The Truth is in There: Improving Reasoning in Language Models with Layer-Selective Rank Reduction”, that is, careful clipping on specific layers of the Transformer model. Branches can significantly improve the performance of the model on certain tasks.

Does fine-tuning large models have to rely on human data? DeepMind: Self-training with feedback is better

Link: https://news.miracleplus.com/share_link/14331

Large Language Models (LLMs) are changing the deep learning landscape, demonstrating superior capabilities in generating human-quality text and solving a variety of language tasks. While the industry has further improved performance on specific tasks through supervised fine-tuning of human-collected data, obtaining high-quality human data faces significant bottlenecks. This is especially true for tasks that involve solving complex problems, requiring significant resources and expertise. How to solve it? Synthetic data generated by models is a promising alternative that can be scalable and cost-effective as long as the quality of the data is maintained. While LLMs are capable of self-evaluating generated data, in this paper, Google DeepMind explores a simpler setup that uses an external scalar feedback signal as a quality indicator for each generated sample.

What are the “Science ChatGPT” in the scientific research field in 2023? What are the explorations of LLM for Science…

Link: https://news.miracleplus.com/share_link/14332

In 2023, the most influential things in the field of artificial intelligence will be GPT-4 and ChatGPT. ChatGPT has single-handedly set off a craze in the field of AI, which has taken the world by storm and seems to have started the fourth industrial revolution. ChatGPT was selected into Nature’s 2023 Top Ten People (Nature’s 10). This is the first time in history that a “computer program”—the first non-human entity—has been selected. “Nature stated that this approach is intended to recognize the role that AI systems that imitate human language play in the development and advancement of science.

Moonshot Al chatbot Kimi Chat updated with new model, support for scanned PDF, etc.

Link: https://news.miracleplus.com/share_link/14333

Domestic artificial intelligence startup Moonshot Al announced updates to its chat robot Kimi Chat, including: 1. Release of the new version moonshot-v1-20231225 model; 2. Improved first-word reply speed; 3. Support for scanned document recognition; 4. Small Program: Kimi smart assistant is online; 5. Light color mode on web page is online; 6. Christmas Easter eggs.

Vincent video software Pika 1.0 is open to all users on the web today

Link: https://news.miracleplus.com/share_link/14334

Early this morning, the Pika team announced on social platform

The throne of the chip industry changes hands: NVIDIA is expected to take advantage of the AI boom to surpass Samsung and Intel in revenue

Link: https://news.miracleplus.com/share_link/14335

As the artificial intelligence craze sweeps across 2023, the semiconductor industry will finally usher in the new king in thirty years – AI chip leader Nvidia; it is predicted that Nvidia’s revenue in fiscal year 2023 is expected to double year-on-year, while Samsung and Intel’s revenue in the same period It is expected to decline year-on-year, and NVIDIA is expected to take away the crown of the chip market.