This article presents a thorough assessment of GPT‑4o’s new image generation features, detailing multiple test scenarios—from simple portrait creation and style transfer to UI design, product rendering, and educational illustrations—comparing its output with Claude‑3.7‑Sonnet, highlighting strengths in realism and weaknesses in Chinese text handling.
The Q‑Eval‑100K dataset, comprising 100 k AIGC images and videos with separate visual‑quality and textual‑consistency annotations, powers the open‑source Q‑Eval‑Score framework that fine‑tunes multimodal models to deliver state‑of‑the‑art, scalable, and objective evaluation—including a “vague‑to‑specific” strategy for long prompts—surpassing existing benchmarks.
When designing AI products, choosing between a Language User Interface—leveraging speech recognition, NLP, and conversational flexibility—and a Graphical User Interface—relying on visual icons, layouts, and intuitive interaction—depends on technology maturity, response speed, and user learning cost, while emerging multimodal designs increasingly blend both for richer, context‑aware experiences.
In 2024 the advertising team deployed an end‑to‑end AIGC pipeline that automatically creates high‑quality ad images, uses the multimodal Reliable Feedback Network and the million‑size RF1M dataset to filter outputs, builds rich offline and online multimodal representations with contrastive and list‑wise learning, and optimizes ranking architecture to deliver scalable, personalized creative selection.
This article explains a step‑by‑step implementation of a multimodal Retrieval‑Augmented Generation system that parses PPT/PDF files, extracts rich text and images with vision models, indexes them in a vector store, and generates answers that combine markdown and relevant slide screenshots.
This tutorial walks through using DataWorks Notebook with GPU support to combine the open‑vocabulary visual model RAM and the language model BERT for zero‑shot multimodal image captioning, covering environment setup, model installation, dataset preparation, tagging code, and result visualization.
Gaode Map and Xi'an Jiaotong University introduce the “Driving by the Rules” task, releasing the MapDR benchmark that integrates lane‑level traffic‑sign regulations into online‑constructed HD maps, and provide modular (VLE‑MEE) and end‑to‑end (RuleVLM) baselines to evaluate rule extraction and lane association.
ByteDance and partners open‑source three AI projects—Goku for high‑quality text‑to‑video generation, Streamer‑Sales for multimodal live‑selling LLMs, and MimicTalk for rapid 3D talking‑head creation—detailing their core features, underlying transformer‑based architectures, training pipelines, and public repositories.
AI video generationTransformerVirtual digital human
The 2024 advertising team introduced a suite of AI‑driven techniques—including a trustworthy feedback network, a large‑scale human‑annotated dataset, multimodal large language model representations, and online ranking architecture upgrades—to dramatically improve the quality, coverage, and personalization of generated ad creatives.
2024 saw the advertising team achieve major breakthroughs in AI-generated ad creatives by introducing a multimodal reliable feedback network to improve image usability, releasing a large human-annotated dataset, and leveraging multimodal large language models for richer representation and more effective online/offline creative selection.
This article recounts a three‑stage interview for a multimodal large‑model position at Taobao's Search AI division, detailing typical questions on CLIP, LoRA, BLIP, Qwen‑VL, Transformer fundamentals, RLHF, and coding challenges, and offers insights on what interviewers focus on.
Google’s Gemma 3, a lightweight open‑source model with up to 27 billion parameters, offers multimodal input, 128K token context, and broad language support, outperforming leading rivals on single‑GPU benchmarks and providing flexible deployment options for developers and researchers alike.
The paper introduces GIFNet, a three‑branch network that leverages low‑level visual tasks and a cross‑fusion gating mechanism to achieve a single, task‑agnostic image‑fusion model with dramatically reduced computation, strong generalization to unseen modalities, and even single‑modal enhancement capabilities.
Taobao’s AIGC pipeline combines a human‑feedback multimodal reward model, audio‑visual joint pre‑training, and Mixture‑of‑Experts distillation to clean data, align outputs with user preferences, and achieve state‑of‑the‑art multimodal LLM performance that drives content cold‑start and conversion gains in e‑commerce.
This article traces the development of AI models—from early word embeddings like Word2Vec and ELMo, through transformer‑based encoders such as BERT and decoder‑only models like GPT‑1/2/3, to recent multimodal systems and scaling laws—explaining their architectures, training methods, and impact on modern AI applications.
Taobao’s new multimodal AI Agent automatically creates high‑quality static and dynamic video covers by planning tasks, consulting a memory of quality criteria, executing frame selection with ReKV streaming and dual‑stage evaluation, generating marketing copy via fine‑tuned Qwen2.5‑7B, and refining layout, resulting in significantly higher click‑through rates, lower latency, and reduced manual effort.
Taobao’s AIGC video generation platform, built on a large‑scale “Faxiang” model that evolved from UNet to DiT, leverages over 2 billion curated e‑commerce videos, expert alignment, Lora fine‑tuning, and multi‑control capabilities to deliver diverse, high‑quality product videos that dramatically boost conversion metrics across the marketplace.
JD Retail’s AI assistant combines a Master‑Sub agent framework, ReAct paradigm, multimodal integration and MoE architecture to improve sales forecasting, pricing, and recommendation accuracy, while the team’s collaborative culture and open talent pathways illustrate how cutting‑edge AI is applied in real‑world e‑commerce.
Transfusion is a 7B‑parameter transformer that jointly trains language modeling and diffusion losses on mixed text‑image data, enabling seamless text generation, image generation, and image understanding within one model and outperforming prior multimodal approaches such as Chameleon across multiple benchmarks.
VARGPT is a novel multimodal large language model that unifies visual understanding and autoregressive image generation within a single architecture, extending LLaVA with next‑token and next‑scale prediction, trained through three staged data‑curated phases and achieving superior performance on numerous vision‑language benchmarks.
This article provides an in‑depth technical analysis of DeepSeek’s model distillation technology, covering its core principles, innovative data‑model fusion strategies, architecture design, training optimizations, performance benchmarks, and the remaining challenges of scaling distillation to multimodal tasks.
The article reviews Google Gemini 2.0 Pro Experimental, detailing its free‑during‑experiment pricing, multimodal understanding, real‑time streaming, native tool integration, usage limits, latency controls, and practical scenarios such as large‑scale code processing and live media handling.
The article presents Ola, an open‑source full‑modal LLM that uses progressive modality alignment to jointly process text, images, video, and audio, and demonstrates competitive performance across image, video, and audio benchmarks, surpassing many specialized models.
The paper introduces MILS, a training‑free multimodal iterative LLM solver that lets large language models perceive and generate across image, video, and audio domains, achieving new state‑of‑the‑art results without any task‑specific data or fine‑tuning.
The article analyzes the 2025 AI large‑model landscape, contrasting slowing foundational breakthroughs with fierce application competition, highlighting MiniMax’s low‑cost linear‑attention models, multimodal advances, and the strategic shift from price wars to sustainable, technology‑driven growth.
The article analyzes the US‑China large‑model race, detailing algorithmic and architectural strengths of OpenAI, Google and Microsoft versus Chinese innovations like Doubao 1.5, MiniMax‑01 and Vidu, and projects a timeline from 2025 to 2033 for China to close the gap.
This article presents a comprehensive overview of the RAG2.0 engine design, covering RAG1.0 limitations, effective chunking methods, accurate retrieval techniques, advanced multimodal processing, hybrid search strategies, database indexing choices, and future directions such as agentic RAG and memory‑enhanced models.
MiniMax, a Shanghai‑based AI unicorn, has open‑sourced its MiniMax‑01 series featuring large‑scale linear attention, secured $600 million in funding, launched multimodal products like Talkie and Hailuo AI, and is positioning itself as a competitive force amid rising geopolitical tensions in the global artificial‑intelligence market.
This issue reviews recent AI industry developments, including Lee Kai‑fu’s clarification on Zero‑One’s strategy, Microsoft’s open‑source Phi‑4 model, the multimodal VITA‑1.5 release, and HaiLuo AI’s advanced Chinese voice‑cloning technology, providing technical details and market implications.
In 2025, large language models will see three key trends—agents becoming pervasive in daily life and industry, the emergence of efficient small models for edge and specialized tasks, and the integration of multimodal capabilities that combine text, images, and audio to enable more natural human‑machine interaction.
This article demonstrates how to create an AI‑driven personal expense‑tracking assistant by leveraging Zhipu's GLM‑4V‑Flash multimodal model for receipt OCR, generating SQL statements, and integrating them with MaxKB workflows and a MySQL database, complete with code snippets and deployment steps.
This article examines the unique multimodal and multi‑granular nature of web pages, compares fusion strategies, proposes a cross‑modal attention approach, outlines fine‑ and coarse‑grained pre‑training tasks, and explores low‑cost adaptor methods for adapting large multimodal models to web‑page modeling in the LLM era.
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 article analyses the rapid growth of Retrieval‑Augmented Generation (RAG) in enterprises, explains why self‑built RAG solutions often collapse under cost and maintenance pressures, and demonstrates how Tencent LeXiang AI Assistant addresses these issues through a robust knowledge‑management core, extensive industry experience, scalable resources, and advanced multimodal capabilities.
Physical Intelligence's new Pi-zero model, built on a vision‑language foundation and fine‑tuned with extensive robot data, outperforms prior baselines across multiple tasks, showcasing the promise of large multimodal foundation models for flexible, robust robot control.
EasyAnimate V5, an Alibaba Cloud AI video generation framework, expands model size to 7B/12B, introduces multimodal control, token‑length based training, and inpaint‑based image‑to‑video strategies, while providing easy deployment via PAI, DSW, and local ComfyUI integration.
This article surveys current unified generation-and-understanding multimodal large-model architectures, compares LLM-centric and LLM-plus-diffusion designs, extracts common insights, details large-scale training tricks from models like Emu3, Chameleon and Janus, and outlines open research directions for visual encoders.
The paper introduces a Multimodal Reliable Feedback Network (RFNet) and a consistency‑condition regularization technique that together boost the usable rate of automatically generated advertisement images while preserving visual quality, supported by a new million‑image annotated dataset and extensive ECCV‑2024 experiments.
Bilibili’s AI‑driven game‑recognition system extracts real‑time LoL events through OCR, hero detection and hot‑spot tagging, generating high‑energy timestamps and interactive overlays that let viewers jump to key moments and view detailed statistics, enhancing spectator engagement and analytical capabilities across major esports tournaments.
Alibaba Cloud’s AI platform PAI recently saw two papers accepted at EMNLP2024—VideoCLIP‑XL, which enhances video‑text representation for long descriptions using a large video‑long‑description dataset and novel pre‑training tasks, and TAPIR, a curriculum‑planning framework that distills instruction‑following abilities of large language models—while also releasing associated models, datasets, and integration tools for users.
The article outlines how to obtain the DataFun Summit e‑book by following the public account instructions and provides concise English summaries of twelve technical sessions covering data lineage, integration, AI language models, multimodal content, game AI agents, lake‑warehouse governance, big‑data architecture, and cluster management.
The CNCC 2024 conference featured a series of expert talks on AI large‑model research, covering paradigm shifts in scientific discovery, knowledge enhancement and governance, data‑infrastructure analytics, vertical‑domain inference, diffusion‑model advances, multimodal model progress, and medical AI applications, illustrating the breadth and impact of large‑model technologies across multiple domains.
The article examines how large language models are reshaping JD Retail's e‑commerce search and recommendation pipelines, detailing industry evolution, technical challenges such as knowledge hallucination, intent understanding, personalization, cost, and safety, and presenting JD's end‑to‑end AIGC architecture, data preprocessing, alignment, evaluation, and next‑generation AI search solutions.
The article reviews the evolution of artificial intelligence from early statistical models through deep learning and Transformer architectures, examines current breakthroughs like multimodal models, and discusses the technical, computational, and safety challenges that must be overcome before achieving artificial superintelligence (ASI).
The talk reviews the evolution of data technologies from early database storage to today’s generative AI-driven era, highlighting how massive data, multimodal processing, and advanced analytics are transforming data systems from cost‑centered infrastructures to value‑focused ecosystems that empower intelligent agents, open data ecosystems, and new application paradigms.
The JD Advertising R&D team applies cutting‑edge AI techniques—including query intent models, multimodal representation pipelines, reinforcement‑learning‑based auction mechanisms, generative recommendation with quantized product tokens, and large‑model infrastructure—to boost traffic valuation, ad relevance, revenue, and creative generation across the platform.
JD's advertising research team combines deep learning, multimodal modeling, reinforcement‑learning auctions, and generative recommendation to boost ad relevance, improve long‑tail product exposure, and overcome large‑model inference challenges in a high‑traffic e‑commerce environment.
The article reviews the rapid advances in generative AI and big‑data technologies, traces the historical development of data infrastructure, and argues that modern data systems are shifting from a cost‑efficiency focus to a value‑centric paradigm driven by multimodal, non‑structured data, vector search and machine‑oriented services.
This article presents JD.com’s multimodal recommendation system architecture, covering content understanding, multimodal ranking and recall models, practical deployment pipelines, and future research directions such as large‑model integration and supply‑side generation, all illustrated with detailed diagrams and Q&A.
This article presents eight practical optimization recommendations for the indexing stage of enterprise‑level Retrieval‑Augmented Generation (RAG) applications, covering chunk creation, abbreviation handling, multimodal document processing, semantic enrichment, metadata usage, alternative index types, and embedding model selection.
This article details Tencent Game's end‑to‑end approach to building intelligent NPCs, covering the opportunities brought by AI, the practical implementation of multimodal LLM‑driven dialogue, knowledge‑augmented retrieval, long‑context handling, safety measures, multimodal expression (voice and facial animation), and system‑level performance optimizations for real‑time deployment.
This article presents a comprehensive technical overview of how large AI models are reshaping financial data production, governance, multimodal document understanding, lakehouse storage, private‑domain model deployment, data‑centric engineering methods, and multi‑agent intelligent advisory within the finance sector.
This article analyzes how Xiaohongshu addresses the decentralization challenge in its recommendation system by strengthening side‑information usage, integrating multimodal signals across the full pipeline, and implementing interest exploration and protection mechanisms, while also outlining future research directions such as generative recommendation and large‑model‑driven user profiling.
From August 11‑16, 2024 the ACL conference in Bangkok featured 14 Ant Group papers covering large‑scale information extraction, decomposed LLMs for semantic search, multimodal hallucination detection, long‑context attention mechanisms, concept‑reasoning datasets, knowledge‑graph alignment, and more, highlighting the group's breadth in natural language processing and AI research.
This article reviews why large language models can be applied to graph learning, outlines their capabilities and graph data characteristics, surveys current research across different graph types and LLM roles, and proposes future research directions for unified cross‑domain graph learning.
Exploring Large Language Models, this article surveys their core concepts, evolution through Transformers, GPT and BERT, generation challenges, diverse applications such as QA, multimodal creation, summarization and retrieval‑augmented generation, prompt‑engineering frameworks and tools, LangChain‑based pipelines, AI‑driven agents, and future prospects toward domain‑specific use, multimodality, and AGI.
The article analyzes the rapid evolution of AI large models—from their role in advancing AGI through massive pre‑training and fine‑tuning, to current market dynamics led by GPT and domestic Chinese players, and finally to future multimodal applications, content‑factory capabilities, and emerging AIGC revenue models projected to reach trillion‑yuan scales by 2030.
The article analyzes current challenges in deploying large AI models, covering robot automation, scaling‑law limits, vertical‑domain use cases, multimodal breakthroughs, algorithmic evolution, and the hardware‑software trade‑offs of training and inference infrastructures, while questioning ROI and practical feasibility.
Vitron is a unified pixel‑level visual multimodal large language model that integrates image, video, and region encoders with a text‑centric strategy, delivering precise pixel‑wise perception and a comprehensive suite of vision tasks from understanding to generation and editing.
This article explains how to handle complex multimodal PDFs in RAG systems, outlines extraction, indexing, and multimodal model integration, details four query‑rewriting strategies (HyDE, stepwise, sub‑question, backward), and presents key evaluation metrics and tools for assessing RAG performance.
The 360LayoutAnalysis project from 360 AI Lab releases lightweight, yolov8‑based layout analysis models covering Chinese and English papers, Chinese research reports, and a general document scenario, providing fast inference, paragraph‑level detection, and open‑source code and weights for flexible document‑understanding pipelines.
This article provides a comprehensive introduction to large language models, covering their historical development, core architecture, training process, prompt engineering techniques, Retrieval‑Augmented Generation, agent frameworks, multimodal capabilities, safety challenges, and future research directions.
The IEEE CVPR2024 conference in Seattle accepted 2,719 papers out of 11,532 submissions, and Ant Group contributed 24 papers covering computer vision, deep learning, digital humans, large models, multimodal remote sensing, vision‑language distillation, federated incremental learning, model‑stealing defense, and more, with one highlighted as a highlight.
This article presents Shopee's comprehensive approach to constructing an e‑commerce knowledge graph, detailing the challenges of heterogeneous data, multi‑language handling, entity disambiguation, and the integration of deep learning and large language models to improve product matching, recommendation, and operational efficiency.
This article presents Shopee's comprehensive exploration of building an e‑commerce knowledge graph, detailing its challenges, construction pipeline, AI‑driven extraction and fusion techniques, multilingual and multimodal modeling, and practical applications ranging from search and recommendation to AI assistants and real‑time updates.
This article documents the end‑to‑end workflow of using large multimodal models and specialized agents to automatically generate a storyboard, compose original music and lyrics, produce keyframes, and assemble a complete music video, while highlighting the remaining manual steps and future automation possibilities.
This article reviews the open‑source 9‑billion‑parameter GLM‑4‑9B model, covering installation, quick‑start inference code, quirky Chinese riddles that highlight its strengths over GPT‑4, extensive benchmark tables for dialogue, multilingual, tool‑calling and multimodal tasks, and its broader impact on the Chinese AI ecosystem.
This article details eBay's practical experience integrating multimodal data and graph neural networks into its recommendation pipeline, covering pain‑point analysis, a twin‑tower multimodal embedding model with triplet loss and TransH, engineering design, experimental results, and key takeaways for future AI‑driven product development.
The Mixed‑Modal Expert Model combines ID features with image and text embeddings through optimized representations and conditional output fusion, dramatically improving advertising recall—especially for long‑tail items—and delivering measurable gains in click‑recall, revenue, CTR, and page views in large‑scale online tests.
This article reviews recent advances in generative recommendation, highlighting challenges such as item representation and multimodal fusion, and summarizing four key research papers that propose novel tokenization, collaborative integration, and transformer-based multimodal approaches to improve recommendation performance.
This article presents OPPO's practical research on deploying multi‑modal pre‑training models across mobile devices and cloud, covering edge image‑text retrieval, text‑image generation and understanding optimizations, and lightweight diffusion model techniques, with detailed algorithmic improvements, performance results, and real‑world application cases.
This article examines OpenAI's GPT‑4o, outlining its key performance, speed, accuracy, latency, multimodal, and resource‑efficiency improvements over GPT‑4, and explains why these enhancements broaden the model's applicability across various AI‑driven applications.
The article introduces 360VL, an open‑source multimodal large language model built on Llama‑3‑70B, describes its novel C‑abs bridge architecture for high‑resolution visual understanding, outlines the two‑stage training with bilingual data, and presents benchmark results showing superior performance over prior LMMs.
The article provides a hands‑on review of OpenAI's newly released GPT‑4o model, covering its multimodal capabilities, real‑time voice demo, desktop client rollout, access options for paid and free users, practical usage tips, and early observations on API performance and limitations.
This article presents Baidu's exploration of multimodal content understanding for commercial advertising, detailing the ViCAN pre‑training model, its contrastive and mask‑language learning tasks, integration across recall, ranking and risk‑control pipelines, quantization with MMDict, and future AIGC‑driven generation, all backed by extensive experiments and Q&A.
Meta’s newly open‑source Llama 3 model pushes the frontier of large language models with a larger context window, Mixture‑of‑Experts architecture, multilingual support, and multimodal capabilities, while facing challenges in transparency, bias, and computational resources, and offering diverse applications from NLU to code generation.
Sora, the newly announced text‑to‑video large model, can generate one‑minute high‑fidelity videos from textual prompts or static images, handling complex scenes, expressive characters, and sophisticated camera motions while also supporting video extension and frame‑filling, positioning it at the forefront of multimodal AI research.
This article reviews Tongyi Lab's work on the OFA framework for generative multimodal pretraining and the ONE-PEACE model for unified multimodal representation learning, detailing their architectures, training strategies, experimental results across vision‑language and audio tasks, and future research directions.
The paper presents MuLTI, a multimodal video‑language model that tackles the memory and efficiency challenges of long video‑text sequences by introducing a Text‑Guided MultiWay Sampler and a Multiple Choice Modeling pre‑training task, achieving state‑of‑the‑art results on video QA and retrieval while drastically reducing GPU memory consumption.
The article details how JD.com leverages domain‑specific and generic knowledge graphs to enhance multimodal product information, improve controlled text generation, and boost LLM performance for e‑commerce copywriting, covering model architecture, copy‑only mechanisms, token‑type encoding, experimental results, and practical deployment scenarios.
OpenAI’s Sora model demonstrates large‑scale text‑conditional video generation using a diffusion transformer that operates on spatiotemporal patches, supporting variable durations, resolutions, and aspect ratios while showcasing emergent simulation abilities, flexible sampling, and multimodal editing capabilities, though it still has notable limitations.
Google’s Gemini 1.5, a new multimodal Mixture‑of‑Experts model, supports up to a million‑token context (10 million internally), can understand text, video, audio and code, learns a new language from a single prompt, and is already being used by Samsung, Jasper and Quora, positioning it as a direct challenger to OpenAI’s flagship models.
In this talk, the speaker examines the dual goals of AI agents—being entertaining and useful—while introducing the concepts of fast and slow thinking, multimodal perception, long‑term memory, retrieval‑augmented generation, and tool integration as essential steps toward building truly valuable digital companions.
This article presents Baidu's commercial multimodal understanding and AIGC innovations, detailing rich‑media multimodal perception, a unified large‑scale representation framework, scenario‑specific fine‑tuning, and practical applications such as marketing copy, digital‑human video, and poster generation.
The article examines the security challenges introduced by large‑model AIGC, outlines three technical upgrade paths—richer training data, few‑shot model fine‑tuning, and multimodal fusion—and demonstrates practical implementations that dramatically improve detection efficiency, accuracy, and scalability.
This article introduces a machine‑learning approach for correlating text and image data, covering preprocessing, feature extraction, model training, experimental results, and future directions, and provides complete Python code examples using NLP and deep‑learning libraries.
Google announced Gemini, a suite of multimodal large language models—including Ultra, Pro, and Nano—that achieve state‑of‑the‑art results on dozens of benchmarks, support native multimodal pre‑training, and are being integrated across Google products such as Bard, Search, and upcoming Pixel devices.
This article presents NetEase Cloud Music's multimodal cold‑start solution, detailing the problem background, feature selection using CLIP, two modeling approaches (I2I2U indirect and U2I DSSM direct), contrastive learning enhancements, interest‑boundary modeling, and evaluation results showing significant gains in user engagement.
This article summarizes a collection of newly released open‑source AI projects covering natural‑language processing, multimodal processing, intelligent agents, recommendation systems, and model training acceleration, providing brief descriptions, key capabilities, and links to their repositories.
OpenAI revealed a new data partnership initiative to collect large‑scale public and private datasets across multiple modalities, aiming to improve AI model safety and usefulness by incorporating diverse, hard‑to‑access human‑generated content while respecting privacy and intent.
This article reviews recent research on augmenting language and multimodal models with external knowledge sources and tool‑calling mechanisms, covering three systems—OREO‑LM for knowledge‑graph reasoning, REVEAL for multi‑source visual‑language pretraining, and AVIS for dynamic tool selection—and their experimental results and implications.
The article recounts a personal test of a multimodal AI agent in Newport Beach and expands into a detailed analysis of current multimodal LLM architectures, memory mechanisms, task planning, tool usage, personality modeling, cost constraints, evaluation challenges, and the broader social and reliability implications of deploying such agents.
This article reviews recent research on augmenting language and vision models by incorporating external knowledge sources such as knowledge graphs, multi‑source retrieval, and dynamic tool‑calling frameworks, presenting three systems—OREO‑LM, REVEAL, and AVIS—and their experimental results.
This article presents MiniGPT-4, a multimodal system that combines a frozen visual encoder (Q‑Former + ViT) with an open‑source large language model (Vicuna), describes its motivation, training pipeline, demo capabilities, observed limitations, and includes a brief Q&A session.
This article reviews the technical background of simultaneous speech translation, compares offline and real‑time scenarios, details ASR and MT technologies, describes the system architecture and design strategies, and discusses the major challenges and solutions for deploying robust, low‑latency translation services.
The article surveys how large language models can be integrated into recommendation systems, either as feature extractors or as end‑to‑end recommenders, showing that LLM‑derived semantics improve recall, ranking, diversity, and user experience, and outlining future multimodal, efficiency, and re‑ranking directions.
This article presents Baidu's commercial multimodal understanding framework and AIGC innovations, detailing rich-media multimodal perception, the VICAN‑12B multimodal representation‑generation model, scenario‑specific fine‑tuning, feature quantization for ranking, and practical applications such as marketing content generation, digital‑human video creation, and poster synthesis.
Recent advances in visual language models enable zero-shot multimodal tasks, and this article explores their application to open-set object detection, prompt learning, and promptable surgical instrument segmentation, highlighting methods like CLIP, CoOp, and the DetPro framework with experimental results across multiple benchmarks.
The CSIG‑hosted "Enterprise Visit – Into Qihoo 360" event on June 29, 2023 gathered over a thousand participants to explore multimodal and cross‑modal learning in the large‑model era, featuring keynote speeches from leading university researchers and Qihoo 360 AI experts, a tour of the company's facilities, and discussions on future AI research directions.