popularization of science

In news about autonomous driving, numbers can easily overshadow explanations. “99%” sounds almost perfect, but without knowing what the denominator is, it has almost no informational value. For a heavy-duty truck designed to operate without a driver in the cab, this distinction is particularly important. October 30, 2026...

Google and Bing have begun managing AI training, search discovery, and answer citations separately. Before webmasters gain more granular control, they need to understand the effective dates, signal boundaries, and the three key considerations for content management.

An AI chip in orbit writing a poem has almost no practical value for the average user; for engineers researching space computing, however, it answers a fundamental question: Can computing hardware on the ground receive missions, run models, and then send the results back in a real space environment? 20...

How far is a machine-generated, lengthy proof from making it into a textbook? The academic collaboration initiative announced by OpenAI on October 26, 2026, has brought this question to the forefront. It involves three areas of research—harmonic analysis, computational complexity, and quantum field theory—and attempts to use various methods of public disclosure to enable the field...

Create a video set at the beach where the viewer walks along the shore, turns around, changes the weather, and continues exploring. While it may seem like just a few extra buttons, there are actually two very different product requirements behind them. The former can tell a complete story within a fixed timeframe, while the latter must constantly respond to the viewer’s choices and ensure that what has already happened influences subsequent scenes...

This adjustment pertains to the training and evaluation of internal state-of-the-art models; it does not eliminate ChatGPT’s ability to access the internet. It brings a fundamental question to the forefront: When AI can use tools and explore alternative paths, what should define the boundaries of its permissions? On October 23, OpenAI ...

On October 22, 2026, Microsoft announced that a study on the diagnostic system MAI-DxO and the sequential diagnostic benchmark SDBench had been officially published in *Science*. Early results from this project had already been released in 2025; what is noteworthy this time is the expanded evaluation, public release...

Introduction: Inside and Outside the Great Wall—A small building nestled among trees in Compound No. 10, Chedaogou, Haidian District, Beijing, is home to the China Ordnance Industry Computer Application Technology Research Institute. On September 14, 1987, an email was sent from here across the Pacific Ocean to Geneva, Switzerland. It contained just one sentence: "Acro...".

When Opus 4.7 was first released, there were a lot of complaints on X. Some people said that one conversation used up her session quota. Some people said that one conversation maxed out her session quota, others said that the cost to run the same piece of code more than doubled from the previous week, and still others revealed that their $200 Max subscription touched down in less than two hours...

What is MCP? What is MCP Model Context Protocol (MCP) is a communication protocol designed for machine learning model services to efficiently manage contextual information during model inference (e.g., session state, environment...

In recent times, the inference model DeepSeek-R1 has arguably been the number one topic in AI. Those who have used it know that the model outputs a piece of thought chain content before outputting the final answer. Doing so improves the accuracy of the final answer. Today's post will take you through the chain of thought...

With the rapid development of artificial intelligence technology, large-scale language modeling (LLM) has become a key force driving this field forward. In order to better master and utilize LLM technology, it is particularly important to understand its core parameters. In this paper, we will take an in-depth look at three key parameters in large-scale language modeling: the Toke...

Overview: Explain what is meant by "inference model" Discuss the advantages and disadvantages of inference models Outline the DeepSeek R1 training methodology Describe the four main approaches to building and improving inference models Share a view of the LLM domain after the release of DeepSeek V3 and R1 Provide a view of the LLM domain at a small cost...

Knowledge Distillation in the Era of Big Models In the field of Artificial Intelligence (AI) and Machine Learning, Knowledge Distillation (KD), as an important model compression technique, is receiving more and more attention in the era of big models now. In simple terms, it works by...

Recently, many people engaged in large model training and reasoning have been discussing the relationship between the number of model parameters and model size. For example, the famous alpaca series LLaMA large model contains LLaMA-7B, LLaMA-13B, LLaMA-33B and LLaMA...

In the field of artificial intelligence and machine learning, especially when building applications such as RAG (Retrieval Augmented Generation) systems and semantic search, efficiently processing and retrieving massive unstructured data becomes crucial. Vector databases have emerged as a core technology to address this challenge. They are not only for storing high-dimensional ...

1. Introduction to AI Agent (LLM Agent) 1.1. Terminology Agent: "Agent" usually refers to a manifestation of intentional action. In the philosophical realm, an Agent can be a human, an animal, or even a concept or entity with autonomy. AI Agent...

In the vast world of Artificial Intelligence (AI), large models play a pivotal role. These large and complex models have shown amazing intelligence and potential through deep learning and massive data training. However, the development path of big models is not a single one, and open source and closed source have become two major trends...

In recent years, China has successively issued relevant regulatory documents on the management of algorithmic recommendation and deep synthesis, as well as the management of generative artificial intelligence services, initially building up a regulatory mechanism for artificial intelligence technologies and services in specific fields. Specifically in the field of generative AI services, the "Generative AI Service Management...

The release of ChatGPT3.5 in December 2022 triggered extensive attention to large models worldwide, and domestic universities took the lead in releasing their products in February 2023, and numerous general-purpose large models and vertical large models emerged in China since June 2023, and the domestic large model industry is now showing a "hundred...