AI Brief
AI的演進
Artificial Intelligence (AI) Evolution


Deep Learning (深度學習)

AI 的應用
CNN
- Image Classification (影像分類)
-
Object Detection (物件偵測)

- Pose Estimation (姿態估計)
- Face Recognition (人臉識別)

GAN
- Image Inpainting (修圖)

- Deep Fake (換臉)

- Stable Diffusion (生成圖片/影片)

RNN
- Text-to-Speech(TTS) : 文字轉語音
- Text-to-Text : Translation (翻譯), Text Generation(文本產生)
- Generative Pretrained Transformers (GPT) : Q&A, Exam (考題問答)
- Large Language Model(LLM): ChatGPT, Gemini, Grok
Dataset: GSM8K (Grade School Math)

AI Competitions & Jobs
Generative AI (生成式人工智慧)
The 55 Best AI Tools for 2025 (Tried and Tested)
LLMs Timeline
AI agents comparison

Microsoft WHAM
Introducing Muse: Our first generative AI model designed for gameplay ideation

Large Language Model (大型語言模型)
Blog: An Opinionated Guide to Using AI Right Now

GPT-5
GPT‑5 的智慧水準全方位大幅提升,從各項學術及人類評量的基準測試中可見一斑,在數學、程式設計、視覺感知與健康領域的表現尤其突出。它在數學 (在 AIME 2025 未使用工具的情況下達到 94.6%)、實際程式設計 (SWE-bench Verified 達 74.9%、Aider-Polyglot 達 88%)、多模態理解 (MMMU 達 84.2%) 和健康 (HealthBench Hard 達 46.2%) 領域的基準測試中,全面刷新最高記錄,而這些進步就體現在日常使用情境中。運用 GPT‑5 Pro 的延伸推理能力,這款模型還在 GPQA 中創下 SOTA 新紀錄,在不使用輔助工具的狀態下取得 88.4% 高分。
Grok-4
- 推理能力大幅提升:AIME 數學、GPQA 科學問答測試表現領先,擅長複雜問題解構。
- 專用編碼模式(Grok 4 Code):現場展示即時編寫並執行 HTML 與 Python 程式。
- 全面多模態互動:支援文字、圖像與語音輸入,新增即時圖像生成。
- 即時資料檢索(RAG 架構):與 X 平台整合,即時獲取新聞、趨勢貼文進行回答。
- Hybrid Transformer-MoE 架構:提升運算效率與任務專業化。
- 超大規模訓練:使用 xAI Colossus 超級電腦,訓練資源達 25 萬顆 Nvidia H100 GPU。
Gemini-3

Introducing the Gemini 2.5 Computer Use model
Prompt: From https://tinyurl.com/pet-care-signup, get all details for any pet with a California residency and add them as a guest in my spa CRM at https://pet-luxe-spa.web.app/. Then, set up a follow up visit appointment with the specialist Anima Lavar for October 10th anytime after 8am. The reason for the visit is the same as their requested treatment.
AIGC (AI Generated Content)
Text-to-Image
- Grok Image 0.9
- Gemini Nano-Banana (Gemini-2.5-flash-image)
Text-to-Video
- OpenAI Sora2
- Nvidia ChronoEdit
AGI - Artificial General Intelligence (通用人工智慧)
AGI stands for Artificial General Intelligence. It’s a theoretical level of AI development where a machine can understand, learn, adapt, and implement knowledge across a wide range of tasks, much like a human being.
Paper: Levels of AGI: Operationalizing Progress on the Path to AGI
Paper: GAIA: a benchmark for General AI Assistants
- 推理型 LLM 的出現,加速了對 AGI 到來的那一天的想像。
- AGI 更像是一種「資源」,而非「工具」
- AGI 將使公司更傾向於裁員並停止招聘新人,因為人類勞動力不再具有經濟價值。
LLM Reasoning
Reinforcement Pre-Training
Microsoft and China AI Research Possible Reinforcement Pre-Training Breakthrough

Nested Learning

AI News
2025
- 2025/11/18 Introducing Gemini 3
- 2025/11/07 Introducing Nested Learning: A new ML paradigm for continual learning
- 2025/10/21 DeepSeek-OCR: Revolutionary Context Compression Through Optical 2D Mapping
- 2025/10/15 releasing Cell2Sentence-Scale 27B (C2S-Scale)
- 2025/10/07 Introducing the Gemini 2.5 Computer Use model
- 2025/10/05 ChronoEdit:Towards Temporal Reasoning for Image Editing and World Simulation
- 2025/09/30 Sora 2 is here
- 2025/09/23 Qwen3-VL:明察、深思、广行
- 2025/09/15 Introducing GPT5-Codex/
- 2025/09/19 xAI Grok 4 Fast
- 2025/09/13 特斯拉發表高精度佔用網路專利!純視覺 AI 強化 FSD 環境建模與停車輔助
- 2025/09/11 Qwen3-Next:迈向更极致的训练推理性价比
- 2025/09/04 Exploring Environments Hub: Your Language Model needs better (open) environments to learn
- 2025/08/28 Introducing gpt-realtime and Realtime API updates for production voice agents
- 2025/08/26 Image editing in Gemini just got a major upgrade
- 2025/08/14 Introducing Gemma 3 270M: The compact model for hyper-efficient AI
- 2025/08/07 Achieving 10,000x training data reduction with high-fidelity labels
- 2025/08/07 Introducing GPT-5
- 2025/08/05 Introducing GPT-OSS
- 2025/07/21 New approach allows drone swarms to autonomously navigate complex environments at high speed
- 2025/07/09 Grok-4
- 2025/06/22 OpenAI底層AGI技術被曝光!前研究主管豪言:從此再無新範式
- 2025/06/10 Microsoft and China AI Research Possible Reinforcement Pre-Training Breakthrough
- 2025/05/25 Gemini 2.5: Our most intelligent AI model
- 2025/05/21 OpenAI Unites With Jony Ive in $6.5 Billion Deal to Create A.I. Devices
- 2025/05/20 Build with Jules, your asynchronous coding agent
- 2025/05/14 AlphaEvolve: A Gemini-powered coding agent for designing advanced algorithms
- 2025/05/05 OpenAI agrees to buy Windsurf for about $3 billion, Bloomberg News reports
- 2025/04/05 The Llama 4 herd: The beginning of a new era of natively multimodal AI innovation
- 2025/03/25 Introducing 4o Image Generation
- 2025/03/25 Gemini 2.5: Our most intelligent AI model
- 2025/03/12 Introducing Gemma 3: The most capable model you can run on a single GPU or TPU
- 2025/02/27 微軟發展自家多模態模型Phi-4-multimodal,56億參數支援裝置端運作
- 2025/02/19 Accelerating scientific breakthroughs with an AI co-scientist
- 2025/02/19 Introducing Muse: Our first generative AI model designed for gameplay ideation
- 2025/02/18 xAI公布號稱世界最強的Grok 3模型、進階功能DeepSearch,新增SuperGrok訂閱方案
- 2025/02/14 聯發創新基地全面開源MediaTek Research Breeze 2多模態基礎模型群,實現繁中AI助理
- 2025/02/11 NXP Acquires AI Chip Startup Kinara
- 2025/02/06 Large Language Models Self-Discover Reasoning Structures
- 2025/02/02 OpenAI Introducing Deep Research
- 2025/01/31 OpenAI o3-mini: Pushing the frontier of cost-effective reasoning
- 2025/01/27 A shocking Chinese AI advancement called DeepSeek is sending US stocks plunging
- 2025/01/23 OpenAI Introduction to Operator & Agents
- 2025/01/21 Trump announces a $500 billion AI infrastructure investment in the US
- 2025/01/20 DeepSeek-R1 Release
- 2025/01/07 17歲高中生寫出「神級Prompt」強化Claude推理能力媲美o1模型,如何實現?
- 2025/01/05 Huggingface smolagents
2024
- 2024/12/31 Introducing smolagents, a simple library to build agents
- 2024/12/26 Introducing DeepSeek-V3
- 2024/12/20 OpenAI Announces ‘o3’ Reasoning Model
- 2024/12/12 Introducing Gemini 2.0: our new AI model for the agentic era
- 2024/12/09 Google DeepMind AI模型GenCast可提供15天氣象預報,比權威機構還準
- 2024/12/07 A New Scaling Paradigm: Meta’s Llama 3.3 70B Challenges “Death of Scaling Law”
- 2024/12/06 OpenAI為期12 天直播活動!揭示全新產品與功能
- 2024/12/05 Generating Worlds
- 2024/12/04 Genie 2: A large-scale foundation world model
- 2024/11/28 Ai2發表全新AI模型OLMo 2 完全開源性能和Llama有得拚
- 2024/11/25 LazyGraphRAG: Setting a new standard for quality and cost
- 2024/10/24 OpenAI Introducing Canvas
- 2024/10/22 Meta’s SAM 2.1 Explained: Smarter Segmentation and Developer Tools For the Future
- 2024/10/22 Introducing Stable Diffusion 3.5
- 2024/10/13 OpenAI unveils experimental ‘Swarm’ framework, igniting debate on AI-driven automation
- 2024/10/12 OpenAI Researchers Introduce MLE-bench
- 2024/10/04 Meta Movie Gen
- 2024/10/02 Blackforest Labs announcing FLUX1.1 pro and the BFL API
- 2024/10/02 OpenAI DevDay2024
- 2024/09/30 Liquid Foundation Models: Our First Series of Generative AI Models
- 2024/09/25 Llama 3.2: Revolutionizing edge AI and vision with open, customizable models
- 2024/09/19 QQwen2.5: 基础模型大派对!
- 2024/09/17 Nvidia NVLM 1.0: Open Frontier-Class Multimodal LLMs
- 2024/09/17 Pixtral 12B - the first-ever multimodal Mistral model.
- 2024/09/12 Introducing OpenAI o1
- 2024/09/05 OpenAI Co-Founder Raises $1 Billion for New Safe AI Startup
- 2024/08/20 Microsoft Unveils Phi-3.5: Powerful AI Models Punch Above Their Weight
- 2024/08/13 Grok-2 Beta Release
- 2024/07/30 NVIDIA Accelerates Humanoid Robotics Development
- 2024/07/29 Introducing SAM 2: The next generation of Meta Segment Anything Model for videos and images
- 2024/07/25 AI achieves silver-medal standard solving International Mathematical Olympiad problems
- 2024/07/23 Introducing Llama 3.1: Our most capable models to date
- 2024/06/25 Etched is Making the Biggest Bet in AI
- 2024/06/18 Google DeepMind, Harvard Develop AI-Powered Virtual Rat to Study Movement
- 2024/06/12 Announcing the Open Release of Stable Diffusion 3 Medium
- 2024/05/21 New models added to the Phi-3 family, available on Microsoft Azure
- 2024/05/13 Hello GPT-4o
- 2024/04/29 TAIDE團隊釋出Llama 3-TAIDE-LX-8B-Chat-Alpha1模型,具臺灣文化的大型語言模型再升級
- 2024/04/18 Introducing Meta Llama 3
- 2024/03/19 TacticAI: an AI assistant for football tactics
- 2024/03/03 California officials give Waymo the green light to expand robotaxis
- 2024/02/29 Figure Raises $675M at $2.6B Valuation and Signs Collaboration Agreement with OpenAI
- 2024/02/27 Apple cancels plans to build an electric car
- 2024/02/18 Sam Altman’s $7 trillion chip dream: Bold vision or delusional fantasy?
- 2024/02/15 OpenAI Sora: Creating video from text
- 2024/02/15 Our next-generation model: Gemini 1.5
- 2024/01/24 Humanoid Robot for Warehouse Use Ready for Mass Production
- 2024/01/17 AlphaGeometry: An Olympiad-level AI system for geometry
- 2023/12/06 Liquid AI, a new MIT spinoff, wants to build an entirely new type of AI
- 2023/12/06 Introducing Gemini: our largest and most capable AI model
- 2023/12/05 Elon Musk’s AI startup — X.AI — files to raise $1 billion in fresh capital
- 2023/12/05 AI Alliance Launches as an International Community of Leading Technology Developers, Researchers, and Adopters Collaborating Together to Advance Open, Safe, Responsible AI
- 2023/11/30 Audiobox: Generating audio from voice and natural language prompts
- 2023/11/29 Millions of new materials discovered with deep learning
- 2023/11/28 Pika, which is building AI tools to generate and edit videos, raises $55M
- 2023/11/21 GAIA: a benchmark for General AI Assistants
- 2023/11/20 AI finds formula for how to predict monster waves by using 700 years’ worth of data
- 2023/11/09 Levels of AGI: Progress on the Path to Artificial General Intelligence from Google DeepMind
- 2023/10/31 DeepMind:A glimpse of the next generation of AlphaFold
- 2023/08/07 GPTBot: OpenAI releases new web crawler
- 2023/07/18 Meta and Microsoft Introduce the Next Generation of Llama
- 2023/02/24 Introducing LLaMA: A foundational, 65-billion-parameter large language model
- 2022/11/30 Introducing ChatGPT
- 2022/09/02 An A.I.-Generated Picture Won an Art Prize. Artists Aren’t Happy.
- 2021/01/05 DALL·E: Creating images from text
AI的影響
-
2022/01/14 AI人工智慧取代大量人力,工業5.0時代它如何影響我們的生活?
根據麻省理工學院(MIT)和波士頓大學(Boston University)經濟學家的一項研究,如果AI技術發展速度加快,到2025年,機器人僅在製造業就可以取代200多萬名工人。 但隨著AI技術變得越來越智能,人工智能技術更有可能會超越人類,甚至可能會讓專業度很高的職業也被淘汰。 -
2023/04/17 80%的工作會被ChatGPT影響!OpenAI研究指出這12種職業最受衝擊,擔心被取代一定要知道
OpenAI與賓州大學這份研究指出,百分之百會被ChatGPT模型取代的職業有:報稅員、網頁與數位介面設計師、作家、數學家、會計師、記者、金融量化分析師、行政助理…。而大部分工作內容會被取代的則有:公關專家、區塊鏈工程師、口筆譯員、排版校對員…。 -
2025/02/19 Accelerating scientific breakthroughs with an AI Co-scientist
AI的未來
Elon Musk latest interview
CES 2025 Jenson Keynote
AI Ascent 2025
長文導讀紅杉資本給創業者的戰略建議:AI 如何成為下一個兆元經濟?
AI Enpowerment (賦能)
Humanoid Robots(人形機器人)
Optimus Gen3
Walker S2
Figure 03
ADAM (DeepMind’s RoboTool)
Paper: ADAM: a robotic companion for enhanced quality of life in aging populations
NVIDIA Isaac GR00T N1
WRC 2025
This site was last updated November 19, 2025.