French AI startup Mistral announced a series of significant updates on September 17, aimed at attracting developers and expanding its market share. The company, valued at $6 billion, has launched new initiatives including a free testing tier, substantial price reductions, and new multimodal capabilities, showcasing its ambition in the increasingly competitive field of AI model providers. Through its API service platform la Plateforme, Mistral introduced the free tier, allowing developers to experiment, evaluate, and prototype at no cost. However, production environment usage may require...

Sequoia Capital partner Pat Grady stated that future billion-dollar AI companies are more likely to emerge from application software rather than foundational model development. Although Sequoia has invested about $150 million in foundational models like OpenAI and xAI, its application layer investments are significantly higher. Grady emphasized the need for AI startups to design products atop models to meet market demands, expressing optimism abo....

The US Department of Commerce has proposed reporting requirements for companies developing cutting-edge AI models and firms providing infrastructure, in response to the Biden administration's executive order on 'safe, reliable, and trustworthy artificial intelligence development and usage.' Key elements of the new regulations include: AI models with over 10^26 integer or floating-point operations must report, with the threshold for biological sequence data models set at 10^23; there is a requirement to disclose model capabilities, information security protections, and red team testing results to ensure AI technologies meet safety standards, protecting against cyberattacks and misuse by non-state actors.

Tesla's founder Elon Musk refuted media reports on social platform X regarding a revenue-sharing agreement between Tesla and his AI company xAI. This agreement was said to enhance Tesla's self-driving software, develop a voice assistant, and software for Tesla's humanoid robot Optimus. Musk stated that Tesla indeed received assistance from interactions with xAI, which accelerated the realization of unsupervised FSD, but there is no need to license anything from xAI. He emphasized that xAI's models are too large to be integrated into Tesla.

Tesla CEO Elon Musk denied reports from The Wall Street Journal regarding a revenue-sharing agreement between Tesla and the artificial intelligence startup xAI. Musk emphasized that while Tesla has gained valuable insights from discussions with xAI engineers that help accelerate the implementation of unsupervised FSD, there is no need to obtain any permissions from xAI. He pointed out that the xAI models are too large to run on Tesla's inference computers. xAI, founded by Musk, aims to compete with OpenAI and plans to use data from its various companies to train models and improve technology.

AnythingLLM is a local AI application developed by Mintplex Labs, designed for everyone to simplify the use of AI technology and lower barriers to entry. It does not require expensive cloud services or complex model deployment and integrates RAG, vector databases, and agent functionalities. It supports open-source models like Llama and Mistral, as well as enterprise-level models such as GPT-4. Suitable for businesses, academia, and the legal field, it provides support for multiple document formats, ensures data privacy, and supports local and cloud-based operations.

On September 2, 2024, Beijing Zhipu Technology Co., Ltd. signed a strategic cooperation agreement with Honor in Beijing to establish a joint laboratory for large AI model technology, marking a new stage in their collaboration in the field of artificial intelligence. Zhipu is committed to developing a new generation of cognitive intelligent large models, focusing on innovation in general-purpose AI foundation models. Honor is a leader in platform-level AI technology in the consumer electronics sector.

Microsoft plans to introduce the 'Smart Media Search' feature in Windows 11, allowing users to search for indexed local video or audio files using voice commands. This feature requires user consent to scan media files on the device and will download and install AI models in the background for transcription and indexing. The feature is still in development, and the AI models may require dedicated neural processing units for calculations. It remains unclear whether users can select specific files for search, raising privacy concerns due to the AI's comprehensive scanning of media content. If selective processing is allowed for users, this feature...

AI startup Magic has completed a $320 million funding round, with participation from Eric Schmidt, Alphabet's CapitalG, and investors like Atlassian, bringing the total funding close to $500 million. Magic focuses on AI model development aimed at generating code and automating software development tasks, featuring an exceptionally long context window of up to 100 million tokens, surpassing current commercial models. Magic plans to collaborate with Google Cloud to build two supercomputers to enhance

Nvidia's latest earnings report shows revenue reaching $30.4 billion, a 122% increase compared to the same period last year, exceeding analysts' expectations. Data center revenue grew by 154%, reaching $26.3 billion, becoming a focal point. Nvidia expects to set a new record for shipments next year, with founder and CEO Jensen Huang mentioning that the company's development is closely tied to the increasing complexity of AI models, aiming to drive an industrial revolution by reducing costs and expanding AI model scales. The significant growth in data center revenue reflects strong demand for AI.

OpenAI's 'Strawberry' project, an AI model focused on solving mathematical and programming problems, was recently showcased in the security domain and is set to launch a chatbot version this fall, possibly integrated into ChatGPT. This model excels in language tasks, especially in solving logical puzzles, and is expected to enhance training quality through synthetic data, reducing 'hallucination' errors for more accurate responses. OpenAI's business is rapidly growing, with a twofold increase in revenue from law degrees and ChatGPT subscriptions year-over-year.

Microsoft has released three new AI models aimed at enhancing its leadership in the multilingual and multimodal fields. The Phi-3.5 Mini Instruct is a lightweight model suitable for task execution, code generation, and logical reasoning, excelling particularly in multilingual and multi-turn conversations. The Phi-3.5 MoE (Mixture of Experts) model is rich in parameters and supports various inference tasks, especially outperforming peers in code, mathematics, and multilingual understanding. The Phi-3.5 Vision Instruct integrates text and image processing capabilities.

The academic community is currently facing a controversy over data usage rights, as many academic publishers begin to package and sell research paper data to technology companies for training AI models. This action has sparked widespread discussion. Deals between Taylor and Francis and Microsoft, as well as Wiley and another company, have led to academic achievements being used to enhance AI systems. This practice has caused waves in the research community. Experts point out that once papers are used for training, it is difficult to remove them later, and the widespread acquisition of datasets enables AI models to learn and generate high-quality text.

Cornell University and other institutions conducted hallucination benchmarking tests on generative AI models like GPT-4o, Claude, and Gemini, finding that these models produce hallucination-free text only about 35% of the time, indicating that the reliability of AI still has room for improvement. The research involved fact-checking through authoritative sources across topics like law, health, and history, designing question sets that included content not covered by Wikipedia. The results showed that OpenAI's models performed the best overall, but the improvement compared to GPT-3.5 was limited. The study pointed out that model size is not related to the frequency of hallucinations.

World Labs, a startup founded by Stanford University AI professor Fei-Fei Li, has completed two rounds of funding in two months, reaching a valuation of over $1 billion and becoming a 'unicorn' company. The latest funding round was led by NEA, with a significant increase in valuation compared to the first round in April. World Labs is dedicated to developing AI models that can accurately estimate the three-dimensional physical properties of real-world objects and environments, aiming for detailed digital replication without the need for large-scale data collection, and is focused on addressing the scarcity of 3D data. The company's business model has not been validated by the market.

Gartner's latest research indicates that by the end of 2025, approximately 30% of generative AI businesses currently under testing will be abandoned by companies due to cost issues. The high deployment costs, ranging from $5 million to $20 million, along with an annual maintenance fee of $11,000, pose major challenges for AI projects. Gartner analyst Rita Sallam states that despite the long return on investment period for AI, 15.8% of companies have seen increases in revenue and 15.2% have experienced reductions in costs after adopting AI.

TikTok pays Microsoft nearly $20 million monthly to access OpenAI's artificial intelligence models via Microsoft's cloud services, bringing approximately $1 billion in annual revenue to Microsoft's cloud division and highlighting the value of AI services in the tech industry. Should TikTok develop its own large language model, it may reduce its reliance on Microsoft's services. OpenAI previously suspended ByteDance's account to investigate whether it violated developer agreements. As OpenAI's exclusive cloud service provider, Microsoft has invested significant resources to support its models, such as ChatGPT.

The EU has initiated a consultation on rules for providers of General AI Models (GPAI), aiming to provide guidance for compliance with the EU AI Act legal obligations and ensure the 'trustworthiness' of GPAI. The consultation is divided into three parts, focusing on transparency, risk classification and assessment, and the review and monitoring of codes of conduct, to establish relevant behavioral guidelines. Participants include GPAI providers, businesses, civil society, rights holders, and academic experts, whose feedback will influence the formulation of the behavioral guidelines, particularly regarding the summary of model training content.

The U.S. Department of Commerce report supports open-weight generative AI models, such as Meta's Llama3.1, believing this will help small companies, researchers, non-profits, and individual developers access AI resources, fostering innovation and competition. This view aligns with that of Federal Trade Commission Chair Lina Khan, who emphasized that open models can promote more small businesses in the creative market. Assistant Secretary Alan Davidson stated that the openness of large AI systems significantly impacts competition, innovation, and risks, and called for monitoring of large AI models broadly.