AI News

This image blends classical Chinese art with modern elements, inspired by the Northern Song Dynasty painter Wang Ximeng's Thousand Miles of Rivers and Mountains. The image shows a magnificent landscape scroll, the green landscape technique makes the rolling hills and rivers vast, with rich color layers and exquisite details. On top of this picturesque view of the rivers and mountains...

With only 14 billion (14B) parameters, Phi-4 demonstrates performance comparable to or even surpassing some larger-scale models through innovative training methods and high-quality data. In this paper, we describe in detail the architecture, features, and training methods of Phi-4, as well as its practical application in ...

Recently, NVIDIA (NVIDIA), in conjunction with MIT and Tsinghua University, launched an open source image generation model called SANA. SANA is not only capable of efficiently generating images with a resolution of up to 4096 x 4096, but also has a very fast generation speed. SANA's performance...

At a time when AI technology is changing rapidly, the cooperation model between tech giants and traditional media is quietly undergoing changes. Recently, OpenAI and the UK's old media, The Guardian Media Group, announced a strategic partnership, and this news has undoubtedly triggered both the AI tech sector and the news and communication...

Would you like to have a podcast auto-played directly from your phone's news feed based on your interests? Google is testing a new feature called Daily Listen...

INTRODUCTION In the field of Artificial Intelligence (AI), fundamental models (e.g., large-scale language models and visual language models) have become a central force driving technological progress. However, it remains a major challenge to effectively improve the generalization ability of these models to adapt to a variety of complex and changing real-world scenarios. Currently, supervised ...

In this era of information explosion, short-video platforms are like a huge "time black hole", devouring a lot of people's fragmented time. In order to combat this digital addiction, various screen health management tools and digital minimalist lifestyles have been created, but the results are often far from satisfactory. Recently, a program called Wiki...

Toward the end of the year, the domestic large modeling field is again spreading good news. Baichuan Intelligence recently intensively released a number of large model products, following the full-scene deep inference model Baichuan-M1-preview and medical augmented open source model Baichuan-M1-14B, and then re-launched the full-modal...

What is Refly? Refly is an AI-native content creation platform built on the concept of "free canvas", which provides users with an AI document editor that is visible and accessible through multi-threaded conversations, knowledgebase integration, contextual memories, intelligent search, and...

The competition between tech giants in the field of artificial intelligence is getting hotter by the day. The latest news shows that Microsoft is secretly developing a series of internal AI models, the performance of which is believed to be enough to compete with industry leaders, including partner OpenAI. An anonymous source familiar with the matter revealed that Microsoft's internal...

Opera, the Norwegian tech company known for its innovative spirit, is once again at the forefront of browser technology. Recently, Opera announced the launch of a new AI agent called Browser Operator, which brings an unprecedented browsing experience to users. This...

Background JPMorgan Chase, the largest bank in the United States, has recently made a major shift in AI adoption. From initially disabling ChatGPT, the company has now fully embraced OpenAI technology with the launch of a generative AI suite called "LLM Suite"...

Recently, the field of large-scale language modeling has been receiving increasing industry attention for a new paradigm of reinforcement learning in the late stages of training. Following the introduction of O-series models such as GPT-4o by OpenAI and the release of DeepSeek-R1, the outstanding performance of the models proves that reinforcement learning in the optimization process...

Imagine a world where tools can write emails, optimize headlines and boost 25% sales - such a future isn't out of reach, it's already here. The evolution of Artificial Intelligence (AI) is reshaping the marketing landscape, providing businesses with cutting-edge tools that can both reduce workloads and increase revenue.AI writing...

Summary Nexa's native inference framework makes the deployment of generative AI models on the device side seamless and efficient. The technology supports a wide range of chipsets including AMD, Qualcomm, Intel, NVIDIA, and homegrown chips, and is compatible with all major operating systems. We provide generative AI ...

Nowadays, the performance of domestic and foreign big models such as DeepSeek is becoming more and more powerful, and the industry generally believes that AI applications will usher in explosive growth in 2025. However, for enterprises, even with powerful big models, they still face the problems of unclear application scenarios and uncertain application forms...

Each of these knowledge points has different content for teachers and students. In 2024, the Massachusetts Institute of Technology (MIT) launched the Day of AI program to fire up the world...

Python in Excel is now officially released for Windows users of Microsoft 365 Business and Enterprise. Last August, Microsoft partnered with Anaconda to integrate Python for Ex...

Yesterday, DeepSeek released a preview of DeepSeek-R1-Lite, an autonomous reasoning large language model that can compete with o1 and show users the full thought process that o1 does not make public. Similar to OpenAI's o1-pr...

Let's start the new year in an exciting way Possibly generated by GPT-5 What if I told you that GPT-5 is real. Not only is it real, but it's already shaping the world in ways you can't see. Here's a hypothetical: OpenAI has developed GPT...

Claude recently announced a revolutionary new API feature, Long Text Cache, which is designed to provide users with a more efficient and cost-effective service experience. The core benefit of this innovative feature is that it allows the Claude model to "remember" long form user input, which includes, but is not limited to,...

A spokesperson for OpenAI has issued an official statement in response to the incident. The statement said that the Sora model is still in the "research preview" phase and that the company is trying to balance creativity with strong security measures so that the technology can be used more widely in the future. ope...

OpenAI o3-mini vs DeepSeek R1: An in-depth comparison of advanced AI inference models to understand the key differences between the two inference models. With the Artificial Intelligence (AI) tech landscape changing rapidly, inference models have become a focal point of technological innovation...

GitHub Copilot is getting a major upgrade: the groundbreaking Agent Mode preview is here, and it's going to disrupt the way you program with AI - instead of passive suggestions, Copilot is evolving to be able to iterate on its own...

The Dify team is excited to announce a major v1.0.0 release update for Dify, the AI application development platform! This milestone release marks a solid step forward in building Dify's next-generation AI app development platform, centered around a new plugin...

Microsoft CEO Satya Nadella announces the end of the app era and a new direction for software development I think the concept of business apps may disappear in the age of intelligent agents. Microsoft CEO Satya Nadella says apps as we know them are disappearing, replaced by intelligent agents...

How to use the o1 properly: don't write prompts; write briefs, focus on the goal: describe what you want, not how you want to get it, and be aware of the o1's pros and cons! Since the release of o1 in October and o1 pro/o3 in December...

After being deeply involved in AI-assisted development for the past few years, I've noticed an interesting phenomenon. While engineers report significant productivity gains from using AI, the actual software we use on a daily basis doesn't seem to be significantly better. What's going on here? I think I know why, and the answer reveals that we...

This document is a PPT of a talk given at Stanford University by Barret Zoph and John Schulman, OpenAI's pre- and post-training leaders (and OpenAI co-founders), sharing their experience in OpenAI's development of Ch...

In today's competitive e-commerce market, how to make your product stand out from the crowd of choices has become a challenge that every brand and business must face. The importance of visual marketing as one of the key factors for e-commerce success cannot be overstated. An attractive and professional product image display not only...

Recently, artificial intelligence company Anthropic has been in the spotlight. Not only has it launched the powerful Claude 3.7 Sonnet model, but it has also made significant progress on the funding front.Mike Krieger, Anthropic's chief product officer...

After 7 months of development, 1 month of testing and 77,376 lines of code, Refly is officially open source! ⚡️🔥🚀 Since the project's inception, Refly has been striving to become a world-class open source project on par with Docker and K8S. Our mission...

I. BACKGROUND AND CHALLENGES With the rapid development of AI technology, large-scale language models (LLMs) have become a core driver in the field of natural language processing. However, training these models requires huge computational resources and time costs, which has led to the rise of Knowledge Distillation (KD) techniques. Knowledge distillation works by ...

Apple Moves Toward a Diverse AI Ecosystem According to the latest findings in the iOS 18.4 beta system code, Apple is reserving the option to integrate Google Gemini for Apple Intelligence. This means that the future of Siri in...

Anthropic, a leading company in the field of artificial intelligence, recently announced the completion of its Series E round of financing, which amounted to $3.5 billion, and the company's post-investment valuation also jumped to $61.5 billion. Against the backdrop of the current complex and volatile global economic situation, such a large amount of financing is undoubtedly...

"A day in the life of an ad optimizer: checking account data at 7am, writing copy at 10am, producing materials at 2pm, adjusting the placement plan at 9pm ......" said an infomercial optimizer in a 2am message: "Party A asked for 50 sets of copy and ...

Guest contributors Lennart Heim and Sihao Huang, this article is cross-posted on Lennart's personal blog, Lennart is a regular contributor to ChinaTalk, and recently participated in a discussion on geopolitics in the era of test-time computing...

Today's official release of Agents in SharePoint, as part of Microsoft 365 Agents, enables users to convert SharePoint sites and documents into scope-limited Agents that are subject matter experts for your business needs. This ...

The o1 series of models are advanced process reasoning models, of which the small-sized o1-mini model has the potential to be stronger than o1-preview in terms of logical reasoning, although its world knowledge capabilities have been cut down. Currently, o1-mini is only open to some free accounts to preview the experience, is your account...

AI-Based Promotional Websites Create a blog that promotes affiliate products. Utilize a generative AI text tool like Jasper or Copy.ai to write tons of blog posts and persuasive content about the products you are promoting. You can also copy and paste text from product sales pages...

Highlights Analytics 1.58-bit FLUX, the first quantization model that reduces the parameters of the FLUX Visual Transformer (totaling 11.9 billion) by 99.5% to 1.58-bit, eliminating the need to rely on image data and drastically reducing storage requirements...

On January 7, a 6.8-magnitude earthquake struck Tingri County, Tibet, and many people are concerned about the progress of the rescue and praying for the safety of the affected areas. Meanwhile, in the midst of people's goodwill and concern, a picture of a "little boy buried under the rubble" quickly became popular on the Internet. This picture was accompanied by the words "Rikaze earthquake", poking the tears of countless people...

It's hard to imagine the amazing changes that would have taken place in the AI space in 2024 if Scaling Law hadn't slowed down, but then again you might be thankful that it's because of Scaling Law's slowdown that it's giving later entrants in the industry a chance to catch up, and giving more ordinary people a ride on this...

In the field of artificial intelligence, innovative company Perplexity has previously launched what it claims to be "the world's first conversational AI search engine," a search engine that can be seen as combining the extensive information retrieval capabilities of Google's search engine with ChatG...

Anthropic has announced that its Claude artificial intelligence (AI) chatbot can now respond in a variety of styles to better match a user's communication preferences, tone and structure. New styles to choose from include formal, concise, and explanatory, in addition to customizable styles. If...

Experimentation is about testing hypotheses. Our hypothesis is that NotebookLM relies on advanced multimodal models like Gemini to provide users with an unprecedented information processing experience. Now, with widespread adoption by millions of users, NotebookLM is officially...

Industry-leading mapping AI comparison, subjective after extensive personal testing Note: All products here are directly available and offer a small daily usage quota. Composition hold is a modeling capability, use requires a technical foundation that may not be supported by the product.

Google has released what it's calling a new "reasoning" AI model - but it's still in the experimental stage, and from the looks of our brief tests, it does have room for improvement. The new model is called Gemini 2.0 Flash Thinking E...

Microsoft recently launched an AI tool called 'Correction', which aims to address the long-standing problem of 'AI illusion' in generative AI models.AI illusion is the problem of errors, inaccuracies, or fictitious information in AI-generated content. This not only makes users question the credibility of AI...

Now that OpenAI has made a major update to the ChatGPT API, we've summarized 8 important points. It is worth mentioning that there is a new "GPT-4-0125-preview" model, which allows for better handling of tasks. 1...

Introduction The field of human-computer interaction is undergoing a profound transformation, and Tavus, a company specializing in the development of advanced AI interaction technologies, has been driving this change. Last year, Tavus launched what was then called the world's fastest conversational video interface (Convers...

At the London AI Workshop Tour, Microsoft made a high-profile announcement of a series of new features for Copilot Studio and teased a public preview of "Autonomous Agents" at the Microsoft Ignite 2024 conference. This ...

Today, DeepSeek, a rising star in China's AI field, has triggered an "earthquake" in the global science and technology industry with its amazing speed and strength. This application, which is known as "the light of domestic AI", not only pushed ChatG...

Still looking for a good AI logo design tool? Just read this article! Logo is not just a simple image, it is the "face" of the brand, which directly affects the brand image and market recognition. Although professional designers can provide high-quality designs, their high costs are often...

The world's first decentrally trained 10B parameter model is born! The Prime Intellect team has announced that they have completed a landmark work: a decentralized training network across the US, Europe and Asia has successfully trained a large 10B parameter model. This...

Recently, the National Supercomputing Internet Platform formally launched a number of large models developed by DeepSeek, including DeepSeek-R1, V3, Coder and other series. Among them, the small version of DeepSeek-R1 provides one-click reasoning service, and users do not need to...

Preface I recently discussed O3 (OpenAI o3) with a few friends, and their reaction can basically be summarized as, "Oh my God, is this really happening?" Yes, it is indeed happening. The next few years are going to be crazy. This is a moment of historic significance, one might even say galactic proportions...

Mistral Small 3: Apache 2.0 Protocol, 81% MMLU, 150 tokens/sec Today, Mistral AI launched Mistral Small 3, a latency-optimized 24 billion...

The Artificial Intelligence (AI) sector continues to attract global investors in 2024. Despite the market's caution about the general economic outlook, the investment boom in the AI sector remains strong. According to analysis, in January 2024, global venture capital funds invested in AI-related companies...

AI technology is penetrating into all industries at an amazing speed, and the field of film and television production is no exception. In this wave of integration of technology and art, the rise of AI has brought unprecedented opportunities for change in the film and television industry. Especially in the field of short drama production, which has long faced high production costs, long production cycles...

01Gamma Task Scenario: quickly do PPT. Applicable people: third-party advertising agencies, individual consultants, corporate internal trainers and other people who need to do a lot of PPT. Recommended reasons: the generated PPT page looks good and has a sense of design; easy to modify, do not need to export multiple versions, a link can be...

On an ordinary weekday morning, a tense atmosphere pervades the company office. The employees were talking in hushed tones, their eyes revealing anxiety. Just then, the CEO of the company made a shocking announcement: the company was going to introduce a brand new "employee" - Copilot Agen...

Dear Friends, The buzz generated by DeepSeek this week has made several important trends clear to many: (i) China is catching up with the US in generative AI, which has significant implications for the AI supply chain; (ii) open weighting models are commoditizing the base model layer...

High-quality AI inference model goes universal. Early this morning, OpenAI released its new inference model o3-mini. OpenAI calls it its most cost-effective inference model, with significantly improved complex inference and dialog capabilities, outperforming its predecessor in science, math, programming, and other areas o...

MediaTek Research recently announced that it has officially open sourced two multimodal models optimized for Traditional Chinese: Llama-Breeze2-3B and Llama-Breeze2-8B, which are optimized for hand...

AI programming tool Cursor has officially released version 0.46, which is a rich update that not only renovates the user interface, but also introduces a new Agent mode and further optimizes Claude 3.7 series....

Highly regarded AI giant OpenAI recently announced the release of a research preview of its latest flagship language model, GPT-4.5, once again generating a lot of attention from the tech community. The highly anticipated new model will initially be available to software developers and ChatGPT Pro...

More than a month has passed since the second wave of updates from Microsoft's launch, and the Copilot Agents intelligences are finally online: Microsoft has directly offered a big move, providing dozens of Copilot Agents intelligences in one breath: Copilot Agents can help...

The Duel of Ocean and Bean: In-depth Analysis of the Characteristics of Conch Ask and Beanbag Conch Ask and Beanbag, two big shots who call the shots in the Internet world, a competition where you sing and I appear, who will be the final winner? I heard that Conch Ask is an online Q&A community launched by MinMax, the technology...

There was this junior high school student who initiated a project in the second semester of his freshman year dedicated to giving his friends a free experience with the GPT-4 model. By the summer of eighth grade, he decided to open source the project and officially named it Chat Nio, while positioning the project as a one-stop ...

In the field of Artificial Intelligence coding, how to make AI intelligences (Agents) utilize tools more effectively to accomplish complex software development tasks has been a core issue of great concern. "Tool Use/Function calling" (TUF) is precisely in the context of this...

Small models can outperform larger models if they are given longer to think. In recent times, there has been an unprecedented amount of enthusiasm in the industry for small models, with a number of 'practical tricks' to allow them to outperform larger scale models in terms of performance. It can be argued that putting the spotlight on improving smaller...

Anthropic is proposing a new standard for connecting AI assistants to systems where data resides. The standard is called Model Context Protocol, or MCP for short. Anthropic says that this today's open source...

In this report, we present you with the top 15 Amazon Product Bonus Markets selected for selling on TikTok in 2025. We selected these opportunities after analyzing the key marketplace factors that drive success: High Demand - Products consistently generate high keyword search volume. Strong...

Driven by the wave of artificial intelligence, speech technology has ushered in unprecedented development opportunities. ElevenLabs, as a technology company focusing on AI speech generation, with its advanced AI technology, successfully transforms text into smooth, natural and highly realistic speech...

Google has come up with a big wave again, yesterday shared the Gemini 2.0 Flash native image generation and editing capabilities, today Gemini has been paid to use the Deep Research tool is now free to use. Many people still don't know...

October 22, 2024 - Microsoft has announced a major update to its Autonomous Intelligentsia feature based on Copilot Studio International Edition, which will be publicly previewed next month. This innovation will provide organizations with powerful tools to enhance the ability of sales, service, finance and supply chain teams...

Before reading the main article, check out DeepSeek R1's self-critique after reading the article 1. on the nature of 'self-evolution' This article keenly captures the core of my design philosophy: freeing ourselves from the shackles of human experience, and autonomously deducing truths from rules and data. AlphaGo's...

(hereinafter referred to as "Wisepac") is concerned that the Bureau of Industry and Security (BIS) of the U.S. Department of Commerce has proposed to add Wisepac and its subsidiaries to the list of export controlled entities. This decision lacks factual basis and we strongly oppose it. Jispectrum is the result of the transformation of Tsinghua University. As...

✨ Little Red Book Nuggets Secrets Revealed! 🔥 Leek programs can be fun? There are also operation tutorials for free! Hello everyone, recently found that everyone is very interested in the small red book related projects, it seems that everyone wants to dig gold in the small red book this blue sea ah! 🚀 Don't worry, today I bring you...

Coze (Button) Automation Work Hands-on Tutorial Introduction In the modern work environment, automation technology is becoming an important tool for organizations to improve productivity with its efficiency, precision and scalability. **Coze (Button)** as a lightweight and highly flexible automation tool for all walks of...

Course Instructor: Dr. Pranav Rajpurkar (Assistant Professor, Harvard University) Course Overview: This course will take you on a deep dive into cutting-edge AI development tools such as PyTorch, Lightning, and Hugging Face, and use VS...

Ideogram, a Canadian AI imaging startup, recently launched its new Canvas feature, which provides users with an unlimited creative canvas designed to organize, generate, edit and combine images. Feature Highlights Unlimited Canvas: users can operate freely in a vast virtual space to generate...

In late 2024, YC Partner Jared predicts that vertical AI Agents will be an emerging market 10x larger than SaaS in the next few years, and that the space could also spawn tech giants with market caps of over $300 billion. By then, Microsoft CEO Saty...

When the phenomenal game "Black Myth: Goku" continues to spark heated debate in the gaming world, and when the DeepSeek big model has become an efficient "code plug-in" in the eyes of programmers, Hangzhou's AI field is once again flooded with innovative forces -- Rokid has launched a new AR Glasses new product, this...

Nvidia (NVDA), Alphabet's Google (GOOGL), and popular startup OpenAI are turning to "synthetic data" factories to meet the massive data requirements for training deep-learning artificial intelligence algorithms. At this week's Consumer Electronics Show, the...

On February 26, 2025, SuperCLUE released the first list of the Project Level Code Generation (SuperCLUE-Project) Measurement Benchmark. For the evaluation program, see: Project Level Code Generation Evaluation Benchmark Release. The evaluation is based on the cooperation of a large model "panel of judges" to evaluate all aspects of...

Baidu Wenxin Intelligent Body Platform has launched a new tool called "Lead Intelligence Body". Lead Intelligence is based on big model technology and is specifically designed to help companies identify, acquire and convert leads more effectively. What is Lead Intelligence? Lead Intelligence is a new tool provided by Baidu Wenshin Intelligent Body Platform for enterprises to...

The gist: Perplexity AI submitted a bid to TikTok's parent company ByteDance on Saturday proposing that Perplexity merge with TikTok's U.S. operations, CNBC has learned. A source familiar with the situation revealed...

Mathematical ability, which encompasses formula derivation, logic chain construction, and abstract thinking, has long been seen as a key area for testing the capabilities of Artificial Intelligence (AI), particularly Large Language Models (LLMs). This is because it does not only test computational power, but also delves deeper into the model's ability to reason, understand, and solve complex problems...

The field of Artificial Intelligence has made impressive progress in recent years, especially in the area of Large Language Modeling (LLM). Many models, such as Qwen, have shown an amazing ability to self-check answers and correct errors. However, not all models are equally capable of self-improvement...

Recently, GenmoAI open-sourced the video generation model mochi 1 preview (10B) with high-fidelity actions and robust cue following capabilities, currently supporting 480p resolution video generation. Today, SiliconCloud, a silicon based flow, went live with an inference accelerated version of mo...

Baidu Wenxin Yiyin: Completely free, opening a new chapter of AI universal Artificial Intelligence field is developing rapidly, as a leader in the field, Baidu has recently announced a major progress again. The much-anticipated Wenxin Yiyan will be available to the public free of charge from April 1,...

Amazon researchers have trained the largest text-to-speech conversion model to date, which is said to exhibit "latent" qualities that enable more natural speech output of complex sentences. This breakthrough may be the key to moving away from unnaturalness in this area of technology. The growth and enhancement of such models has been...

Recently, a well-known novel platform's new agreement triggered a furor. The agreement makes it clear that the platform will use the works of signed authors for AI training, triggering strong dissatisfaction among many netizen authors. Some authors called for the suspension of work, and even launched a "break initiative". Some authors complained to the relevant departments about the platform's copyright problems...

OpenAI recently announced the launch of its NextGenAI program, offering $50 million in funding and access to its newest modeling API. The program brings together 15 of the world's top universities and research institutions to advance artificial intelligence (AI) in education...

ORIGINAL: A Study of Programming Competition Competencies Based on Large Reasoning Models The following is a brief summary for ease of reading. 1. INTRODUCTION 1.1 BACKGROUND AND MOTIVATION In recent years, Large Language Models (LLMs) have made significant progress in program generation and complex reasoning tasks. Programming competitions, especially like the International ...

Today Anthropic announced the release of an upgraded version of Claude 3.5 Sonnet and a brand new model, Claude 3.5 Haiku.This update not only improves encoding capabilities, but also introduces a groundbreaking feature, Computer Use, which is now available in the public test...

According to reports, "Runxin" by the Chongqing University Party Committee, led by the Ministry of Education, the introduction of the wisdom of the spectrum of education industry, the ability of large language models of digital ideological education products. "Runxin" adopts "intelligent body + general large model + industry-specific model + public knowledge base + subject expert knowledge base fusion of the number of intelligent people" mode of construction, with ...