Earlier this year, Google released the sixth generation of TPU, the Trillium, which is the most powerful TPU to date. Today, Trillium is officially available for Google Cloud customers. Google has trained its latest AI model, Gemini 2.0, using the Trillium TPU, making it the strongest AI model created by Google so far. Now, enterprises and startups can leverage the same robust, efficient, and sustainable infrastructure. The core of the AI supercomputer: Tri

Large Language Models (LLMs) like GPT and Llama have sparked a revolution in the field of artificial intelligence, but efficiently training these massive models while aligning them with human values remains a challenge. Reinforcement Learning with Human Feedback (RLHF) has become an important training method for LLMs in recent years, but traditional RLHF frameworks have limitations in flexibility, efficiency, and scalability. To address these issues, the ByteDance Doubao Large Model Team has open-sourced an RLHF framework called HybridFlow.

Recently, ByteDance officially addressed rumors regarding an 'intern disrupting large model training'. The company confirmed that an intern did maliciously interfere with the model training tasks of the commercialization technology team's research project. However, this action did not affect the company's official commercial projects and online business, nor did it involve other large model businesses of ByteDance. Moreover, ByteDance pointed out that claims circulating online about 'involving over 8,000 nodes and causing losses of over ten million dollars' are seriously exaggerated. According to public reports, the intern was part of ByteDance's commercialization technology team.

The social networking platform X (formerly Twitter) updated its privacy policy on Wednesday, sparking widespread attention and discussion. The new policy indicates that X will, by default, allow third-party "partners" to use platform data to train AI models unless users actively choose to opt out. This move not only highlights X's efforts to seek new revenue sources but also raises concerns regarding user data privacy protection. Key changes in the new policy include: Third-party data usage: New terms allow third-party partners to use X user data, including for AI model training. Users have the option to choose.

Social network X (formerly Twitter) updated its privacy policy on Wednesday, indicating that it will allow third-party 'partners' to use X data to train their AI models unless users opt out. This change may position X similarly to other social media platforms and media organizations that consider licensing data to AI companies as a potential new revenue source. In the updated privacy policy, X added a paragraph detailing how X user data will be used and how users can opt out.

At a recent video cloud technology conference, Volcano Engine unveiled a significant innovation: a video preprocessing solution for large model training. This technology has been successfully applied to the Doubao video generation model, marking a significant advancement in AI video generation technology. Volcano Engine's President, Tan Dai, emphasized that AIGC and multimodal technology are profoundly changing user experience. Leveraging experience from Douyin, Volcano Engine is actively exploring the integration of AI large models with video technology to provide comprehensive solutions for enterprises. Wang Yue, Head of Video Architecture at Douyin Group, pointed out that large model training faces numerous challenges.

With the rapid development of artificial intelligence technology, a new high-paying profession has quietly emerged - AI Trainers. The rapid growth of this industry not only reflects advances in AI technology but also reveals the challenges faced by current AI models and future development directions. In the early stages of AI development, model training mainly relied on low-cost workers to perform basic data labeling tasks. However, with the intensification of AI competition and the increasing complexity of models, the demand for high-quality AI trainers has surged. These trainers are no longer limited to undergraduates but include practicing physicians.

At the 2024 Volcano Engine AI Innovation Roadshow, Volcano Engine President Tan Dai unveiled the latest Doubao · Video Generation Model. This model incorporates several advanced technologies, including the efficient DIT Fusion Computing Unit, which enables deep compression encoding of video and text; it also employs a novel diffusion model training approach to ensure consistency in generating multi-camera videos; in addition, the model integrates a deeply optimized Transformer architecture, significantly enhancing the generalization ability of video generation.

Fluxgym is a new training tool aimed at AI enthusiasts, designed to lower the barriers to model training. It has extremely low hardware requirements, needing only 12GB of VRAM to conduct AI model training, significantly expanding the participant base. By combining a simple front-end with a powerful back-end, Fluxgym provides an easy-to-use interface, allowing users to initiate the training process in just three simple steps. Its core advantages lie in support for low-spec devices, a simplified user interface, and a smooth training process, enabling AI newcomers to get started quickly.

The release of PaddlePaddle Framework 3.0 features a core upgrade to unified automatic parallelism technology, aimed at simplifying the distributed training process for large models and improving development efficiency. The new version supports mixed parallelism techniques from four-dimensional to five-dimensional, employing various parallel methods such as data parallelism, tensor model parallelism, pipeline parallelism, and grouped parameter slicing parallelism, effectively enhancing the training efficiency of large models. The automatic parallelism technology uses tensor slicing syntax to automatically infer distributed slicing states and add communication operators, reducing the difficulty of development. The principles of automatic parallelism include distributed tensor representation and slicing inference.

NVIDIA has open-sourced two new large models, Nemotron-4-Minitron-4B and Nemotron-4-Minitron-8B, which utilize structured pruning and knowledge distillation for efficient training. This significantly reduces training requirements, minimizing data and computing power consumption. Compared to traditional methods, the new models reduce training token data by 40 times and save 1.8 times in computing costs. By optimizing Llama-3.18B, structured pruning simplifies the model structure, while knowledge distillation enhances performance.

According to the CEO of Anthropic, the current cost of training AI models is as high as $10 billion, and it could rise to $100 billion or even $1 trillion in the next three years. This forecast has sparked concerns about whether the AI bubble is about to burst.Image Source Note: The image was generated by AI, provided by the image authorization service provider Midjourney Dario Amodei points out that as AI models continue to evolve, hardware demand will also grow exponentially, becoming a major

The Brazilian National Data Protection Authority (ANPD) issued a "precautionary measure" prohibiting Meta, Inc. from using its citizens' data to train AI algorithms. This decision immediately suspended the changes made to Meta's privacy policy last month, which added the company's AI training to its data handling disclosures.The ANPD stated: "Following an initial analysis, considering the serious and irreparable risks to users, the agency has temporarily ordered the suspension of the privacy pol

Tencent recently released the 2.0 version of its independently developed Xingmai Network, a significant upgrade that brings a substantial performance boost to large-scale artificial intelligence model training. The new version has achieved breakthroughs in multiple aspects, including network scale, hardware performance, communication protocols, and fault diagnosis. It is reported that in terms of network scale, Xingmai Network 2.0 supports networking of 100,000 card groups in a single cluster, p

Adobe, a name well-esteemed in the creative industry, has earned the reputation of being a "Copyright Guardian" for its stance on copyright protection. However, the company has recently been caught in a storm of public opinion due to a quietly updated service term. This February, Adobe silently updated its product service terms, adding a striking new clause: users must agree that Adobe can access their works automatically and manually, including those protected by confidentiality agreements. Ado

Tencent Cloud recently launched an upgraded version of the Xing Mai Network 2.0, aiming to enhance the efficiency of large-scale model training. In the previous version, the synchronization communication time for the computation results of large models accounted for over 50% of the total, resulting in low efficiency. The new version of Xing Mai Network 2.0 has been upgraded in several aspects:1. Supports networking of up to 100,000 cards in a single cluster, doubling the scale, with a 60% improv

Recently, the Beijing Internet Court held online hearings for four copyright infringement cases filed by illustrators against the developers and operators of AI art software. The plaintiffs in these cases are all renowned illustrators who discovered that their original works, published on a content-sharing platform, were extensively mimicked by the AI art software and used as training material to generate new artworks. These AI-generated artworks not only closely resemble the style of the plaintiffs' works but are also used heavily for commercial purposes, severely harming the interests of the plaintiffs.