CoTyle introduces a novel open‑source framework that generates unique image styles from a numeric style code, eliminating the need for reference images, lengthy prompts, or LoRA modules, and demonstrates superior style consistency compared to existing solutions like Midjourney.
The article analyzes how generative AI is reshaping enterprise software into AI‑as‑Labor, expands TAM, outlines moat considerations, and presents a greenfield GTM playbook that helps inexperienced founders avoid super‑star traps and build fast‑iterating, high‑margin AI‑native products.
This article explains the over‑optimization problem in GRPO‑based flow models, analyzes why importance‑ratio clipping fails, and introduces GRPO‑Guard with RatioNorm and cross‑step gradient balancing, showing through extensive experiments that it stabilizes training and improves image quality across multiple diffusion backbones and tasks.
The paper identifies imbalanced concept distributions as the main obstacle to arbitrary concept‑combination in text‑to‑image/video generation, proposes the token‑level IMBA Distance and a lightweight IMBA Loss that adaptively re‑weights training tokens, and demonstrates through extensive experiments and a new Inert‑CompBench benchmark that this loss dramatically improves compositional ability without extra data.
Tim Berners‑Lee warns that the rise of generative AI and large language models is reducing web traffic, jeopardizing the ad‑driven internet economy and creating a feedback loop where AI both consumes and generates online content, threatening the sustainability of digital advertising.
This article introduces Spacetime‑GR, a large‑scale generative recommendation model that integrates hierarchical geographic POI indexing and spatiotemporal token encoding to enhance POI prediction for Amap, detailing its pre‑training pipeline, data cleaning, curriculum learning strategy, experimental results, scaling law observations, and the resulting improvements in hit rate and discovery rate.
With the rapid rise of generative AI audio models, this article explores the fundamentals, key metrics, and the “impossible triangle” of imperceptibility, robustness, and capacity in audio watermarking, and presents practical implementations such as SynthID and AudioSeal that embed and detect invisible watermarks for secure AIGC provenance.
BoltzGen, an all‑atom generative model released by MIT and collaborators, unifies protein folding and binder design with a geometric continuous representation and a flexible design language, training on multimodal datasets and demonstrating nM‑level affinity for 66% of 26 diverse targets including proteins, nanobodies, peptides and small molecules.
Generative AI tools like DALL·E, Midjourney and deep‑fake platforms are enabling criminals to create highly realistic images, videos and documents, leading to a surge in sophisticated insurance fraud across auto, property, health and life lines, and forcing insurers to overhaul detection and regulatory practices.
MIT’s SpectroGen model incorporates physical priors into a variational auto‑encoder to transform a single‑modality spectrum into high‑fidelity cross‑modal spectra, achieving up to 99% correlation with experimental data and surpassing traditional methods in accuracy, as demonstrated on IR‑Raman and XRD‑Raman tasks using the RRUFF database.
World Labs unveiled RTFM, a real‑time frame model that runs on a single H100 GPU, generating persistent, interactive 3D worlds from 2D images without explicit 3D representations, highlighting the growing computational demands of generative world models and their potential to reshape AI-driven spatial intelligence.
This article details Baidu's generative recommendation ad recall framework, introducing the COBRA system and its three development stages—dense representation compression, sparse quantization with ID generation, and dense‑sparse cascading—highlighting coarse‑to‑fine inference, performance gains, long‑sequence extensions, online deployment, and future research directions.
This technical article provides a comprehensive comparison of diffusion models and flow matching, covering their intuitive explanations, underlying mathematics, training objectives, sampling efficiency, theoretical guarantees, practical examples, and code implementations to illustrate how each generative approach works.
RFdiffusion2 introduces a novel deep generative approach that eliminates residue enumeration and sequence indexing, enabling atom‑level protein backbone generation from simple chemical reaction descriptions, achieving a 100% success rate across 41 benchmark cases and providing a step‑by‑step demo on the OpenBayes platform.
This article reviews the YOLO-Count model, a fully differentiable, open‑vocabulary object counting system that guides text‑to‑image generators to produce the exact number of objects specified in prompts, achieving state‑of‑the‑art results on both generic counting and controlled image synthesis tasks.
The DataFun GenAI Summit 2025 brings together leading experts from Alibaba, Tencent, Ant Financial, and other tech giants to showcase the latest breakthroughs in large-model research, generative AI, multimodal understanding, and real-world deployments across finance, e-commerce, media, and enterprise services.
This article explores how the convergence of business intelligence and large‑model generative AI creates new business value, outlines the evolution of BI technology through three stages, and details Baidu’s ChatBI design, implementation challenges, and real‑world impact.
McKinsey’s 2025 report outlines a future HR model driven by generative AI that delivers hyper‑personalized employee experiences, frictionless organization structures, and a stronger human‑centric leadership approach, showing how technology can boost productivity, satisfaction and financial performance across enterprises.
This article, excerpted from Baidu’s Data Platform technical salon, explores how generative AI reshapes Business Intelligence by outlining three perspectives: the technical trend and business value, the design principles of Baidu’s ChatBI platform, and practical challenges and solutions encountered during its deployment.
This article walks through a complete, low‑cost SEO pipeline built on Amazon Bedrock’s generative AI, covering automated keyword extraction, search‑intent analysis, product‑vector matching with FAISS, and AI‑driven content creation within a serverless, event‑driven architecture that scales efficiently while reducing manual effort.
This article explains how Siemens harnesses generative AI to create an integrated enterprise knowledge chatbot, detailing the challenges of fragmented data, the “XiaoYu” solution, its ecosystem, and the broader impact of AI on productivity and innovation across the company.
This article explains Baidu's generative ranking model GRAB, detailing how scaling laws from large language models inspire a new recommendation paradigm, the model's architecture, custom attention mechanisms, training strategies, deployment optimizations, and the resulting business gains in CTR and revenue.
YOLO-Count introduces a fully differentiable, open‑vocabulary object counting model that guides text‑to‑image generators to produce the exact number of objects specified in prompts, achieving state‑of‑the‑art performance on both generic counting and controlled image synthesis tasks.
Amazon Finance built an AI assistant that combines Amazon Bedrock, Claude 3 Sonnet, and Amazon Kendra to let analysts query financial data in natural language, achieving a 30% reduction in search time, an 80% increase in accuracy, and high precision and recall across data‑discovery and document‑search tasks.
AniME introduces a director‑driven multi‑agent system that combines a custom model‑selection protocol (MCP) with the open‑source AniSora V3 model to automatically generate consistent, high‑quality long‑form animation from story scripts, handling everything from storyboard creation to video editing and quality evaluation.
Recent MIT research demonstrates that a generative AI model trained on massive antibacterial datasets can design millions of novel molecules, with several candidates showing potent activity in mouse models, highlighting AI’s shift from discovery to de novo drug design.
Drawing from a recent AI hardware lecture, this article outlines the market opportunities, product definition frameworks, development processes, and the evolving role of product managers, illustrating concepts with real-world examples like AI companion toys and self‑distillation models.
The article argues that generative AI will replace traditional record‑based test automation, dramatically boost productivity, expand coverage beyond human imagination, and force QA engineers to shift from script writing to AI‑driven strategy and supervision.
Tencent’s newly released Yan system advances interactive video generation by delivering high‑fidelity, real‑time, editable content for games, virtual worlds and AIGC, featuring a three‑module architecture—Yan‑Sim for AAA‑level simulation, Yan‑Gen for multimodal generation, and Yan‑Edit for granular editing—while also introducing a large‑scale high‑quality dataset and efficient inference optimizations.
This article distills McKinsey’s latest report on Agentic AI, explaining why widespread generative AI adoption has delivered limited economic impact, how AI agents can unlock scalable value across vertical use‑cases, and what CEOs must do to lead the transition to an agent‑centric enterprise.
Microsoft Research’s AI for Science team released the open‑source BioEmu model, a generative diffusion architecture that leverages AlphaFold’s Evoformer and extensive MD and stability data to efficiently sample protein conformational ensembles, achieving near‑MD accuracy in free‑energy and mutation stability predictions while dramatically reducing computational cost.
The paper introduces SADA (Stability‑guided Adaptive Diffusion Acceleration), a novel paradigm that dynamically allocates sparsity per token using a unified stability criterion, enabling efficient ODE‑based sampling for diffusion and flow‑matching models, achieving up to 1.8× speedup with negligible fidelity loss across SD‑2, SDXL, Flux, ControlNet and MusicLDM.
FixrLeak combines AST analysis with generative AI to detect and automatically repair Java resource leaks, dramatically reducing manual effort, improving code quality, and demonstrating the transformative potential of AI‑driven solutions in large‑scale software engineering.
The article summarizes Professor Long Mingsheng’s presentation on the Timer series of time‑series large models, detailing the three core challenges of industrial time‑series analysis, the evolution from statistical methods to generative models, and the technical breakthroughs of Timer 1.0, 2.0 and 3.0 that enable multi‑task, long‑context, and trillion‑scale forecasting for industrial digital transformation.
The LaVin-DiT paper introduces a large‑scale vision diffusion transformer that combines a spatiotemporal variational auto‑encoder, a joint diffusion transformer with full‑sequence joint attention, and 3D rotary position encoding to enable unified, efficient generation across diverse visual tasks such as segmentation and video prediction.
McKinsey’s latest State of AI report reveals that 78% of surveyed firms now use AI in at least one business function, with generative AI adoption rising to 71%, and highlights that CEO governance, workflow redesign, and risk mitigation are critical drivers of measurable EBIT impact, especially for organizations with revenues over $5 billion.
Prof. Dou Dejing shares his journey into Fudan University's Data Intelligence Lab, outlines the history and synergy of big data and AI, reviews generative AI breakthroughs, evaluates large‑model strengths and weaknesses, and explores their expanding industrial applications and market potential.
This article explains what high‑bandwidth memory (HBM) is, outlines its brief history, compares it with DDR, LPDDR and GDDR, describes why large language models and generative AI drive its demand, and reviews its architecture, PCB requirements, market status, and future outlook.
This investigative article traces Scale AI’s evolution from a MIT‑dropout’s data‑annotation startup to a $29 billion AI infrastructure leader, detailing its founder Alexandr Wang, core products, government contracts, competitive advantages, and the strategic shift toward defense‑focused AI solutions.
This article details JD Retail Technology’s AI‑driven virtual try‑on system that combines a 12B Flux‑Fill diffusion model with a high‑quality virtual model library and integrates with the JingMai A/B testing platform, cutting production costs to zero, slashing cycle time to half a day, and increasing order conversion rates by over 80% during the 618 shopping festival.
Gartner's 2025 China AI Top Ten Trends report outlines ten key directions—including open‑source GenAI models, self‑built strategies, agent AI, frugal solutions, engineering focus, collaborative defense, talent growth, ubiquitous adoption, inclusive ecosystems, and data‑driven AI—while providing forecasts through 2030.
This article examines the rapid evolution of artificial intelligence—from Generative AI to Agentic AI—and explains how enterprises can adopt AI‑Native development models, address full‑stack challenges, upgrade data infrastructure, and leverage Chat BI and LangStudio platforms to create intelligent, data‑driven applications.
This guide explains how to design AI‑generated business cards that look like code editor windows by using a detailed prompt template, compares model performance (4o, iDream, Doubao), and offers practical tips for handling Chinese characters and formatting.
This guide walks you through installing and configuring Dify—a open‑source LLM application platform—on your local machine using Docker, integrating it with Ollama for custom models, and exploring its core features such as chat assistants, agents, workflows, and tool extensions, all illustrated with step‑by‑step screenshots and code snippets.
Join Baidu senior algorithm engineer Ji Zhi at the DataFun Summit 2025 to explore how generative AI transforms ad recommendation recall, covering item representation, evolving solution architectures, long‑sequence challenges, and practical insights for building efficient large‑model recommendation systems.
OneRec, an end-to-end generative recommendation system from Kuaishou, uses an encoder-decoder architecture, reward-based preference alignment, and reinforcement learning to dramatically improve video recommendation efficiency, boosting user engagement and reducing operational costs while achieving scaling-law performance comparable to large language models.
AI is rapidly transforming digital design and development by enabling hyper‑personalized, predictive user experiences, generative UI creation, emotion‑aware interfaces, seamless multi‑device interactions, ethical design practices, AI‑driven 3D asset generation, and dynamic data visualizations, reshaping how designers and developers build tomorrow’s products.
The Interactive Intelligence Lab of Ant Technology Research Institute presented 21 accepted CVPR 2025 papers covering visual generation, editing, 3D vision, digital humans and multimodal AI, highlighting tools such as MagicQuill, Lumos, Aurora, FLARE, LeviTor, MangaNinja, AniDoc, Mimir, AvatarArtist, DiffListener, MotionStone, TensorialGaussianAvatars, DualTalk, CompreCap and Uni-AD.
Snowflake unveiled a suite of AI agents and integration tools at its 2025 summit, promising tighter, more reliable AI/ML workflows, natural‑language data access, and rapid, no‑code data integration that can cut implementation time by up to 90% for enterprises.
This article explains the fundamental principles of diffusion models, using physics and machine‑learning analogies to describe forward and reverse diffusion, the role of Gaussian noise, iteration trade‑offs, U‑Net architecture, and shared‑weight training for image generation.
Recent large language models have shown strong reasoning abilities, and this work extends chain‑of‑thought reasoning to autoregressive image generation by introducing T2I‑R1, a dual‑level (Semantic‑CoT and Token‑CoT) framework trained with reinforcement learning that unifies high‑level planning and low‑level token generation, achieving state‑of‑the‑art results.
This article introduces EchoMimic V1 and V2, open‑source generative digital‑human systems that turn a single voice clip and a portrait photo into synchronized talking avatars, covering their technical background, architecture, training strategies, performance comparisons, and potential application scenarios.
This article introduces Spring AI, a Java‑based AI framework inspired by LangChain, explains its core capabilities such as model management, inference, extension and Spring ecosystem integration, and provides step‑by‑step installation, configuration, sample code, and guidance for creating MCP services.
JD’s advertising platform replaces rule‑based recall with a generative large‑model pipeline that unifies e‑commerce knowledge, multimodal user intent, and semantic IDs across recall, coarse‑ranking, fine‑ranking and creative optimization, while meeting sub‑100 ms latency and sub‑¥1‑per‑million‑token cost through quantization, parallelism, caching, and joint generative‑discriminative inference, delivering double‑digit performance gains and paving the way for domain‑specific foundation models.
Google Cloud Next 25 unveiled a wealth of AI advancements, including five new generative models, a groundbreaking Agent‑to‑Agent protocol, upgraded AI‑powered developer tools, expanded AI applications across Workspace, and the high‑performance Ironwood TPU for inference, offering developers a clear view of the latest AI landscape.
The article examines how cryptographic watermarking can secure AI‑generated content by embedding robust, undetectable, and unforgeable signatures, reviews existing methods such as SynthID and Video Seal, and discusses recent research using pseudo‑random codes, error‑correcting codes, and diffusion inversion to improve watermark resilience.
The article defines Agentic AI as autonomous, goal‑directed systems that can act and learn from experience, contrasts it with Generative AI’s passive, single‑step content generation, and illustrates the practical advantage of Agentic workflows through Andrew Ng’s HumanEval benchmark where a step‑wise approach outperforms zero‑shot prompting even for older models.
This article explains how recent MIT research uses stochastic differential equations to model diffusion and flow processes, defines training objectives, explores conditional guidance, compares U‑Net and diffusion transformers, addresses memory challenges with latent diffusion, and surveys applications ranging from robotics to protein design.
The article defines AI, AGI, and AIGC, outlines their core characteristics, functions, and typical application scenarios, compares their scopes, explains how AIGC is an AI application while AGI represents a broader, still‑theoretical goal, and summarizes their interrelationships.
The article examines how generative AI will automate many coding tasks yet create new opportunities for software engineers, emphasizing the need for human supervision, ethical oversight, and advanced roles such as AI integration, system architecture, and cybersecurity in the evolving tech landscape.
The UltraFusion HDR algorithm combines generative AI with traditional exposure fusion to recover details and produce natural‑looking high‑dynamic‑range images even when the exposure gap reaches up to 9 EV, turning over‑exposed or under‑exposed shots into high‑quality photos.
This guide shows how to quickly integrate Alibaba Cloud's AI Search platform and the DeepSeek‑R1 large model with Elasticsearch to create a smart AI Assistant that automates cluster diagnostics, query generation, and visual analytics for operations tasks.
With the rise of LLM‑based generative AI, this article examines the emerging GenBI (Generative Business Intelligence) paradigm, detailing why self‑serving analytics are needed, the progress of Text‑to‑SQL, an LLM‑driven agent architecture, practical AWS Bedrock implementation, technical choices, lessons learned, and future outlook.
The UltraFusion HDR algorithm, developed by Shanghai AI Lab with CUHK and Zhejiang University, combines generative AI with exposure fusion to recover detail and balance lighting even when exposure differences reach up to 9 EV, enabling high‑quality images from ordinary cameras without hardware upgrades.
This issue of the weekly AI digest explores Xiaohongshu’s NoteLLM recommendation system, compares Chinese text generation in video AI across major platforms, highlights Alibaba’s Tongyi Wanxiang breakthroughs, discusses Keras founder François Chollet’s new AGI‑focused lab, and reviews Google’s Veo 2 and Imagen‑3 advancements.
The article provides a detailed analysis of Google’s AI Agent whitepaper, explaining the agent architecture, core components such as models, tools, and orchestration layers, comparing extensions, functions, and data stores, and demonstrating practical implementations with LangChain and Vertex AI to illustrate how targeted learning can boost agent performance.
Redis has unveiled a multi‑threaded query engine that dramatically increases query throughput and lowers latency for vector similarity searches, offering up to 16× performance gains and enabling real‑time Retrieval‑Augmented Generation (RAG) workloads in generative AI applications.
The 2024 Generative Large Model Security Assessment White Paper, jointly authored by the Chinese Academy of Sciences, the Ministry of Public Security's Third Research Institute, and Ant Group's Ant Security Lab, was unveiled at the inaugural CCF China Data Conference, offering a comprehensive review of model risks, ethical concerns, and evaluation methods to guide research, industry practice, and policy making.
This article explains what AIGC agents are, outlines their four main characteristics, describes the underlying transformer‑based architecture, dual‑stage learning, probabilistic generation and feedback optimization, and explores their current and future commercial use cases across content creation, knowledge bases, customer service, internal operations, and product design.
The Kubernetes Community Day held in Jakarta on November 30, 2024 featured Alibaba Cloud experts presenting best‑practice sessions on scaling generative AI workloads, eRDMA network acceleration, container image security, and OpenTelemetry‑based observability within the ACK Kubernetes platform.
The 2024 CNNIC Generative AI Report reveals that 2.3 billion Chinese users (16.4% of the population) are engaging with AI tools, with Baidu’s Wenxin Yiyan leading the market at 11.5% share and dialogue‑based applications capturing 62% of usage.
This article examines why scalability is a core requirement for AIGC products and presents a comprehensive engineering architecture—including modular design, distributed systems, resource scheduling, queue management, and layered architecture—to achieve high performance, cost efficiency, and long‑term maintainability.
This article presents a KDD‑2024 accepted solution that replaces autoregressive generators with a non‑autoregressive model for video recommendation reranking, detailing the challenges, model architecture, novel loss function, extensive offline and online experiments, and practical Q&A from the conference.
This article examines the AIGC engineering architecture, detailing its data, model, fine‑tuning, inference, application, and monitoring layers, and explains the distinct responsibilities and challenges of application engineers, algorithm engineers, and “alchemy” specialists, highlighting how this structured approach accelerates generative AI productization.
The paper proposes a multimodal Reliable Feedback Network (RFNet) and a consistency‑regularized fine‑tuning method (RFFT) that dramatically increase the proportion of usable advertising images generated by diffusion models while preserving visual appeal, and introduces the large‑scale RF1M dataset for training and evaluation.
The 2024 GitHub Octoverse report reveals that Python has surpassed JavaScript to become the most popular language on the platform, driven by the surge in generative AI, data‑science workloads, and a rapid increase in public AI projects and developer diversity.
At the 2024 Baidu World conference, CEO Li Yanhong presented a one‑hour keynote outlining Baidu’s recent breakthroughs in large‑model technology—including the retrieval‑augmented iRAG system, the rise of intelligent agents, and the launch of the no‑code “Miaoda” tool—highlighting their industrial applications and future impact.
Linus Torvalds, creator of the Linux kernel, argues that most generative AI marketing is empty hype, claiming only 10% is real, and says he will largely ignore AI until its practical value is proven, while noting industry investment and mixed opinions from experts.
This survey acts as a comprehensive portal that organizes AIGC research across seven domains—text, image, and audio generation, cross‑modal association, text‑guided image and audio synthesis, and supporting resources—detailing seminal models such as GPT, Diffusion, CLIP, DALL·E, Stable Diffusion, MusicLM, and key papers that shaped each field.
This article describes Siemens' journey in applying generative AI and Retrieval‑Augmented Generation to create an internal knowledge chatbot, detailing the business challenges, technical architecture, data integration, multi‑modal capabilities, deployment outcomes, and strategic lessons for enterprise AI adoption.
This article examines the capabilities and limits of generative AI coding assistants, outlines three interaction modes, proposes methods to cut verification costs, and suggests workflow redesigns that let developers boost productivity while maintaining code quality.
This guide walks you through setting up Ollama for local large‑model serving, creating a Spring Boot project with Spring AI support, writing a unit test to query the model, and explains how to add the necessary dependencies for AI integration in existing Java applications.
This article introduces the Six Thinking Hats framework, explains its benefits, describes AI agent concepts and platforms, and provides a step‑by‑step guide with prompt examples for creating a low‑cost, fully‑featured Six Thinking Hats assistant using generative AI tools.
Sequoia’s third Generative AI Act report argues that foundation models have plateaued, the focus is shifting to inference‑time reasoning, with OpenAI’s o1 model exemplifying a new scaling law that makes inference compute the key driver of future AI breakthroughs.
In this interview, Denodo executives discuss the origins, challenges, and future of data fabric and data virtualization, explore how generative AI and retrieval‑augmented generation enhance data management, share customer success stories, and offer strategic insights for enterprises navigating digital transformation.
This article explores how digital twins evolve from design tools to real‑time monitoring platforms, how integrating generative AI and retrieval‑augmented generation (RAG) boosts AI accuracy and situational awareness, and why software teams must adopt these combined technologies to stay ahead in modern operations.
At the 2024 Cloud Expo, Alibaba Cloud’s CEO Wu Yongming highlighted the rapid evolution of generative AI over the past 22 months, emphasizing its shift from mobile apps to reshaping both digital and physical realms, the exponential drop in inference costs, and the accelerating demand for AI‑driven computing infrastructure.
The article explains how AI databases combine structured, unstructured, and vector data, integrate machine‑learning, NLP, and generative models, and why platforms like Vespa are emerging as open‑source solutions to meet the performance and scalability demands of modern generative AI applications.
Microsoft Research’s AutoGen Studio offers a low‑code web and Python interface built on the open‑source AutoGen framework, enabling developers to quickly prototype, enhance, and combine AI agents into complex workflows while providing drag‑and‑drop design, debugging tools, and Azure integration for secure, scalable multi‑agent applications.
IBM China abruptly removed access for more than a thousand research and testing staff, signaling a sudden, large‑scale layoff that coincides with the company's aggressive push into generative AI, cloud services, and a broader global restructuring effort.
This article presents six hands‑on Python generative‑AI projects—ranging from a Llama 2 chatbot built with Streamlit and Replicate to natural‑language‑to‑SQL conversion using LlamaIndex and SQLAlchemy—complete with environment setup, required code snippets, deployment tips and resource links for further exploration.
The 2024 AI‑powered PPT market has surged to nearly ten million users, with Baidu Wenku capturing about 80% of the share, while competitors like WPS AI PPT and Ai PPT lag behind, driven by superior multi‑modal generation, higher quality layout, and extensive content resources.
Redis has launched a multi‑threaded query engine that vertically scales its in‑memory database, dramatically increasing query throughput and lowering latency for vector similarity searches, thereby addressing the performance demands of real‑time retrieval‑augmented generation in generative AI applications.
The article examines the ultra‑large‑scale, high‑bandwidth, low‑latency, and automated network infrastructure needed for training generative AI models, covering custom network designs, congestion control, deterministic RDMA, topology choices such as Fat‑Tree, and emerging deterministic networking technologies.
Galileo’s Hallucination Index evaluates 22 leading generative AI models using a contextual‑adherence metric, ranking Claude 3.5 Sonnet as the overall RAG leader, Gemini 1.5 Flash as the most cost‑effective, and highlighting open‑source and context‑length performance nuances for AI practitioners.
In 2024 generative AI has moved from isolated coding assistance to a core, team‑wide engine that augments every stage of software development, integrates with low‑code platforms like NetEase’s Tango, and powers specialized tools such as Copilot, Project IDX, and LangBase to boost efficiency, customization, and code quality across enterprises.
This article explains how the scaling law—massive data, compute, and a simple transformer architecture—drives generative AI breakthroughs, how Tencent applied this principle to build larger ad models and the "Hunyuan" large model, and how advertising systems must evolve to truly understand content and users.
The article explains how NeurIPS ranks among top AI conferences, introduces the newly selected “Auto‑Bidding in Large‑Scale Auctions” competition, outlines its technical background, four generations of bidding strategies—from classic control to generative models—and details the competition’s tracks, rewards, and how researchers can participate.
EasyAnimate v3, an open‑source video generation system from Alibaba Cloud AI Platform, introduces Diffusion Transformer‑based architecture, Hybrid Motion Module, and Slice VAE to enable image‑to‑video, text‑to‑video, and unlimited‑length video creation with up to 720p/144 fps resolution on modest GPU memory.
Kuaishou has officially open-sourced Kolors, a state-of-the-art text-to-image diffusion model that leverages ChatGLM3 for advanced bilingual text understanding and employs a two-stage training strategy to achieve photographic image quality rivaling leading proprietary systems.
