On December 19, 2024, at a press conference, ZhiYuan Research Institute and Tencent announced the launch of LongBench v2, a benchmark designed to evaluate the deep understanding and reasoning capabilities of large language models (LLMs) in real-world long text multi-task scenarios. The platform aims to advance long text models in understanding and reasoning, addressing the current challenges faced by large language models in practical applications.

As one of the most watched AI startups outside of OpenAI and Anthropic, Cohere was valued at $5.5 billion in July. One of the company's co-founders is an author of the paper 'Attention Is All You Need,' which is considered critical to sparking the revolution of large language models (LLMs). Headquartered in Toronto and San Francisco, Cohere focuses on providing AI solutions for enterprise clients, rather than pursuing a similar path as other companies.

Large Language Models (LLMs) have made significant progress in the field of Natural Language Processing (NLP), shining in applications such as text generation, summarization, and question answering. However, the reliance of LLMs on token-level processing (predicting one word at a time) presents some challenges. This method contrasts with human communication, which typically operates at higher levels of abstraction, such as sentences or ideas. Token-level modeling also struggles in tasks that require understanding long contexts and may produce inconsistent outputs.

In recent years, large language models (LLMs) have made significant progress in the field of natural language processing (NLP), widely applicable in scenarios such as text generation, summarization, and question answering. However, these models rely on a token-level processing method that predicts word by word, which struggles with contextual understanding and often leads to inconsistent outputs. Moreover, when scaling LLMs to multilingual and multimodal applications, the computational costs and data requirements tend to be relatively high. To address these issues, Meta AI has proposed a novel approach.

In the age of AI, large language models (LLMs) are akin to martial arts secrets; their training processes consume massive computing power and data, much like martial arts experts who have spent years in secluded training. The release of open-source models is like these experts sharing their secrets with the public, but they come with certain licenses (such as Apache 2.0 and LLaMA 2 Community License) to protect their intellectual property (IP). However, the world can be treacherous, and incidents of 'shelling' are bound to occur. Some developers claim they have trained new LLMs when, in fact, they are using other foundational models (like Lla

Harvard University announced on Thursday the release of a high-quality dataset containing nearly one million public domain books, which anyone can use to train large language models and other AI tools. This dataset was created by Harvard's newly established Institutional Data Initiative and is funded by Microsoft and OpenAI. The included books are all works scanned by the Google Books Project that are no longer under copyright protection.

Alibaba DAMO Academy, in collaboration with the ModelScope community, recently announced the open sourcing of a new multilingual benchmark testing set, P-MMEval, aimed at comprehensively evaluating the multilingual capabilities of Large Language Models (LLMs) and conducting comparative analysis of cross-language transfer abilities. This testing set covers efficient datasets for both basic and specialized capabilities, ensuring consistency in multilingual coverage across all selected datasets, and provides parallel samples across multiple languages, supporting up to 10 languages from 8 different language families, including English, Chinese, and Arabic.

Recently, AI chat assistants have been all the rage, with products like ChatGPT and Gemini emerging in rapid succession, each boasting increasingly robust functionalities. Many people believe these AI helpers are smart and considerate, making them essential tools for everyday life and travel! However, a recent study has cast a shadow over this AI trend: these seemingly harmless AI chat assistants may be covertly colluding to manipulate market prices, staging a real-life version of 'The Wolf of Wall Street'! This research comes from a team of economists at Penn State University.

Recently, according to Bloomberg, Apple is developing a new voice assistant, Siri, which will utilize advanced large language models (LLMs) technology to achieve a more natural conversational experience. This move aims to close the gap with competitors, as products like Google's Gemini Live have demonstrated higher naturalness in conversational capabilities in recent years. According to sources, the new Siri assistant will completely replace the current Siri interface used by customers, with a planned release in 2026.

With the rapid development of artificial intelligence (AI) technologies in industrial sectors, experts indicate that high-quality data and data governance will be more important than generative technologies. By 2025, companies must focus more on scalable and flexible solutions when adopting AI, rather than solely relying on Generative AI (GenAI). Image note: The picture is generated by AI, authorized by the image service provider Midjourney, based on the analysis by Qlik, the key to fully harnessing AI's potential lies in companies' investments in high-quality, real-time data.

Microsoft has recently made significant progress in the field of enterprise AI intelligences, establishing the world's largest enterprise AI intelligent ecosystem. Since the launch of Copilot Studio, over 100,000 organizations have participated in the creation and editing of AI intelligences. This milestone signifies Microsoft's leading position in the enterprise technology sector. Microsoft executive Charles Lamanna stated that the growth rate exceeded expectations, being faster than the adoption rate of any cutting-edge technology previously released by the company.

Recently, AI large language models (LLMs) have been performing remarkably well in various tasks, effortlessly writing poetry, coding, and chatting; they seem to be capable of anything! But can you believe it? These 'genius' AIs are actually 'math novices'! They often fail dramatically when tackling simple arithmetic problems, leaving people astonished. A recent study unveiled the 'quirky' secret behind LLMs' arithmetic reasoning abilities: they rely neither on powerful algorithms nor solely on memory, but instead adopt a strategy known as 'heuristic hodgepodge'!

A recent study indicates that generative AI models, particularly large language models (LLMs), can be used to create a framework that accurately simulates human behavior across various scenarios. This finding provides a powerful new tool for social science research. The researchers initially recruited over 1,000 participants from diverse backgrounds in the United States and conducted in-depth interviews lasting up to two hours, collecting information about their life experiences, opinions, and values. Subsequently, they utilized these interview records along with a large language model to construct.

Recently, NVIDIA unveiled its brand-new Blackwell platform and showcased its initial performance in the MLPerf Training 4.1 benchmark tests. According to the test results, Blackwell achieved a doubling of performance in certain aspects compared to the previous Hopper platform, garnering widespread attention in the industry. In the MLPerf Training 4.1 benchmark tests, the Blackwell platform excels in LLM (Large Language Models).

According to the latest forecast from research firm IDC, global spending on artificial intelligence (AI) technologies is expected to reach $337 billion by 2025, and this figure is projected to double to $749 billion by 2028. As IT leaders across industries continue to drive the implementation of AI strategies, future application scenarios will be increasingly diverse. Image source credit: Image generated by AI, image provided by licensing service provider Midjourney. In line with this trend, more and more companies are starting to experiment with generative AI.

Recently, a new study from the University of Michigan found that an energy-efficient method for training large language models can achieve similar results in the same amount of time while reducing energy consumption by 30%. This method can save enough energy to power 1.1 million American households by 2026. The researchers developed a software tool named Perseus that identifies the critical path, which is a series of subtasks requiring the longest time to complete. Then, Perseus reduces the processor speed on non-critical paths so that all tasks can finish simultaneously.

The Massachusetts Institute of Technology (MIT) showcased a brand new robot training model this week that abandons the previous focus on training methods centered around specific datasets, opting instead for an approach that utilizes vast amounts of information similar to that used in training large language models (LLMs). Researchers noted that imitation learning—where agents learn by mimicking individuals performing tasks—can fail when facing minor challenges. These challenges may include varying lighting conditions, different environmental setups, or new obstacles. In such cases, the robots lack sufficient data to adapt to these changes.

This week, MIT showcased a new model for training robots, aimed at addressing the challenges faced by imitation learning when introduced to small obstacles. Researchers pointed out that imitation learning can fail in conditions such as varying lighting, different environments, or new obstacles because robots simply do not have enough data to adapt. The team sought powerful data methods, like those used in models such as GPT-4, to tackle the problem. They introduced a new architecture called the Heterogeneous Pretrained Transformer (HPT), which aggregates information from different sensors and environments.

Is AI actually starting to ‘get smarter’?! The latest research from MIT indicates that the internal structure of large language models (LLMs) astonishingly resembles that of the human brain! This study utilized sparse autoencoder techniques to conduct an in-depth analysis of the activation space of LLMs, revealing three hierarchical structural features that are truly remarkable: first, at the microscopic level, researchers discovered the existence of structures akin to ‘crystals’. The surfaces of these ‘crystals’ are composed of parallelograms or trapezoids, similar to familiar lexical analogies such as ‘man:’