"Conquered by You Just Like This" | 4 Million People, Blasted by AI Until They Crumbled, Laughed Through Tears, and Couldn't Stop Coming Back

🖼️ OpenAI Opens DALL-E 3 to Free Users

🧠 Diffusion-Augmented Agent (DAAG)

🍎 Apple Open-Sources New Image Generation Model ml-mdm

🎙️ Amazon Music Launches AI-Powered Topics Feature

🔢 Alibaba Cloud Qwen2-Math: A New Generation of Math Models

🚀 Kylin Operating System AI Edition and Intelligent Computing Edition Officially Released

🎨 Baidu Netdisk AI Photo Editing

🌐 360 Enterprise Security Browser Upgraded

💰 OpenAI Fund Leads Opal Investment

📈 Metis Technology Valued at $150 Million

🎭 "Toxic-Tongued AI" Goes Live, Roasts 4 Million People to Tears

OpenAI Opens DALL-E 3 to Free Users

On August 9, 2024, OpenAI announced some good news: free ChatGPT users can now generate images using the DALL-E 3 model, with a daily limit of two images. Once the quota is exceeded, ChatGPT prompts users to upgrade to ChatGPT Plus or try again the next day.

The model, launched last September, was initially available only to paid ChatGPT Plus subscribers. In February, OpenAI announced that DALL-E 3 would add C2PA watermarks to generated images to help users identify AI-created content. In April, OpenAI introduced a new editing interface for DALL-E 3, allowing users to make fine-tuned adjustments to generated images based on descriptions. Users can access DALL-E image generation through ChatGPT on web, iOS, and Android platforms.

Diffusion-Augmented Agent (DAAG)

Imperial College London and Google DeepMind jointly introduced the "Diffusion-Augmented Agent" (DAAG), an innovative framework that leverages large language models (LLM), vision language models (VLM), and diffusion models (DM) to improve sample efficiency and transfer learning capabilities of embodied agents in reinforcement learning.

By transforming videos through diffusion models to align with target instructions, DAAG relabels agents' past experiences. No human supervision is required, making it suitable for lifelong learning scenarios. Reward detector learning, experience transfer, and new task acquisition capabilities are all enhanced.

Apple Open-Sources New Image Generation Model ml-mdm

Recently, Apple introduced a new image and video generation technology called "Matryoshka Diffusion Models" (MDM). This technology features a unique "nested diffusion model" architecture that enables seamless image generation from low resolution to high definition.

MDM can process images at different resolutions simultaneously, achieving high-quality image generation with rich detail and strong realism. The "NestedUNet" core architecture reinforces the "matryoshka" concept, with each level containing smaller yet fully functional sub-structures, improving learning efficiency and generation quality. MDM employs a progressive training method from low to high resolution, enabling more stable and efficient performance in high-resolution image generation. MDM is expected to play a significant role in virtual reality, augmented reality, film production, game development, and other fields.

Amazon Music Launches AI-Powered Topics Feature

Amazon Music recently announced an innovative AI-powered feature called Topics. It analyzes podcast transcripts and descriptions to identify key themes. AI-generated "topic" tag buttons are reviewed by human moderators before publication. Users can click any tag to get a list of podcast episodes related to that theme. Currently, the Topics feature is limited to users in the United States on iOS or Android operating systems.

Alibaba Cloud Qwen2-Math: A New Generation of Math Models

On August 8, 2024, the Qwen team announced a new addition to the Qwen2 open-source family — the Qwen2-Math series. These models are specifically designed for mathematical problem-solving and include 1.5B, 7B, and 72B versions. Built on the Qwen2 large language model (LLM), they are pre-trained on specialized mathematical corpora.

The Qwen2-Math base model is pre-trained on large-scale high-quality mathematical resources including web text, books, code, and exam questions, demonstrating mathematical capabilities that surpass existing open-source and closed-source models such as GPT-4o. The Qwen2-Math-72B-Instruct model further improves performance through instruction fine-tuning that combines reward models with binary signals for correct answers.

Kylin Operating System AI Edition and Intelligent Computing Edition

At the 2024 China Operating System Industry Conference, the Kylin Desktop Operating System (AI Edition) and Kylin Intelligent Computing Operating System were officially unveiled.

The new Kylin Desktop Operating System supports advanced features including AI search, image generation, AI Q&A, and text-to-code. By integrating AI technology, it provides a smarter and more convenient user experience. The Kylin Intelligent Computing Operating System is designed specifically for intelligent computing centers to meet high-performance computing demands. Optimized for intelligent computing center application scenarios, it enhances computing efficiency and data processing capabilities.

Baidu Netdisk AI Photo Editing

Baidu Netdisk AI Photo Editing is a product launched by Baidu that integrates multiple AI technologies. Applying Baidu AI technology, it automatically performs efficient and precise photo editing.

Baidu's Cloud Engine technology leverages 9 major portrait analysis and detection capabilities, 86 portrait and image beautification techniques, and over 1,000 visual technology patents to deliver personalized AI photo editing services. Portrait analysis and detection accurately identifies facial features, providing foundational data support for editing. Image beautification techniques include but are not limited to skin smoothing, face reshaping, background reconstruction, and clothing wrinkle removal — comprehensively enhancing photo quality. Over 1,000 patented visual technologies ensure the professionalism and innovation of the editing service.

360 Enterprise Security Browser Upgraded

The 360 Enterprise Security Browser has been fully upgraded with integrated AI technology, providing enterprises with a smarter and more secure office environment. The upgraded browser supports AI search, office assistant, and AI app store features, aimed at improving enterprise work efficiency and data security.

AI empowerment through integrated 360 AI search engine, AI office assistant, and AI app store significantly boosts work efficiency. Multi-layer protection measures covering browser-native security, web data security, and user behavior security ensure enterprise information security. A unified access portal simplifies configuration processes and strengthens app security and quality management.

OpenAI Fund Leads Opal Investment

The OpenAI Fund recently led a $60 million Series B round for hardware startup Opal. Follow-on investors include YouTuber Casey Neistat, TikTok brothers Charli and Dixie D'Amelio, as well as Founders Fund and Kindred Ventures.

Founded in 2020 by Veeraj Chugh and Stefan Sohlstrom, formerly of Uber and Google, Opal is headquartered in California and develops professional webcams, having released a new software application called "Opal Composer." Following this funding round, Opal plans to shift some resources toward developing other AI-powered electronic devices, described as consumer-facing "creative tools."

Metis Technology Valued at $150 Million

Metis Technology recently announced the completion of a new funding round exceeding 100 million RMB, led by Ant Group with Luminous Ventures as a co-investor, bringing its post-money valuation to $150 million. In its early angel round, Metis Technology received investment from Future Capital and Siman Technology, though the specific amount was not disclosed. In May 2021, Metis Technology completed a Pre-A round of tens of millions of RMB, with funds primarily used for product iteration and promotion.

Shanghai Metis Network Technology Co., Ltd. was founded in April 2018, focusing on natural language processing (NLP) and AI technology R&D, with a user base approaching ten million. The company develops products in AI search, AI writing, and legal translation directions, including Metis AI Search, Metis Writing Cat, Metis Translation, and MetaLaw. The founding team brings diverse backgrounds spanning artificial intelligence, computer engineering, and law. Core team members include Chairman and CEO Kerui Min, technology expert Yue Tang, and Chief Operating Officer Yiwei Wang.

Wordware's "Toxic-Tongued AI" Goes Live

"Twitter Personality" [1] is an AI Agent built on Wordware [2]. It quickly went viral for its sharp and humorous roasting capabilities, reaching over 4 million users in 11 days, with peak traffic hitting 106,000 visitors per hour.

"Twitter Personality" can analyze users' social media content and deliver personalized roasts. Targets have included Elon Musk, Taylor Swift, and others — with the AI's roasts proving precise and funny. Users worldwide have been inputting their own social media accounts, eager for the AI's take. Netizens have responded enthusiastically to Wordware's roasts, jokingly quoting "conquered by you just like this" — some finding it hilarious, others saying it hit too close to home.

"I do not expect my life to go smoothly, but I hope that when I encounter life's difficulties, I can be their match. Even alone, I have the courage to stand alone. Later, in the depths of winter, I finally learned that within me lay an invincible summer."

— Notebooks, 1942–1951 (Carnets)

Editorial Team

Editor: Yuki

Design: Ivan

For business partnerships, add on WeChat: Rwkfbcianvd

References

[1] "Twitter Personality": https://twitter.wordware.ai/

[2] Wordware: https://www.wordware.ai/