Recently, Google announced that its AI-based fuzz testing tool OSS-Fuzz successfully identified 26 vulnerabilities in open source code repositories, including a medium-severity vulnerability in the OpenSSL cryptographic library. Google's open source security team stated in a shared blog post: 'The discovery of these vulnerabilities marks a new milestone in automated vulnerability detection: each vulnerability was found through AI-generated and enhanced fuzz testing targets.' Image source note: Image generated by AI, image licensed from Midjour.

Recently, xAI engineer Hieu Pham posted an 'explosive' message on social media claiming that xAI's large language model Grok3 had successfully proven the Riemann Hypothesis, leading to a halt in the model's training. This news quickly sparked heated discussions in the AI community, with netizens expressing shock and struggling to discern the truth. After all, the Riemann Hypothesis is one of the seven Millennium Prize Problems, hailed as the 'crown jewel' of conjectures, its difficulty is well understood. Hours later, Pham revealed in another post that it was just a joke.

The rapid development of deep learning relies on the scale of datasets, models, and computational power. In the fields of natural language processing and computer vision, researchers have already discovered a power-law relationship between model performance and data scale. However, in the field of robotics, especially in robot manipulation, such scalable patterns have yet to be established. A research team from Tsinghua University recently published a paper exploring data scaling laws in robot imitation learning and proposed an efficient data collection strategy that collected sufficient data in just one afternoon, enabling...

Recently, a new study from the University of Michigan found that an energy-efficient method for training large language models can achieve similar results in the same amount of time while reducing energy consumption by 30%. This method can save enough energy to power 1.1 million American households by 2026. The researchers developed a software tool named Perseus that identifies the critical path, which is a series of subtasks requiring the longest time to complete. Then, Perseus reduces the processor speed on non-critical paths so that all tasks can finish simultaneously.

Is AI actually starting to ‘get smarter’?! The latest research from MIT indicates that the internal structure of large language models (LLMs) astonishingly resembles that of the human brain! This study utilized sparse autoencoder techniques to conduct an in-depth analysis of the activation space of LLMs, revealing three hierarchical structural features that are truly remarkable: first, at the microscopic level, researchers discovered the existence of structures akin to ‘crystals’. The surfaces of these ‘crystals’ are composed of parallelograms or trapezoids, similar to familiar lexical analogies such as ‘man:’

In the world of artificial intelligence, the reasoning capabilities of machine learning models, particularly large language models (LLMs), have been a focal point for scientists. Recently, Apple's AI research team published a paper titled 'Understanding the Limitations of Large Language Models in Mathematical Reasoning,' shedding light on these models' limitations when addressing logical problems. In the paper, researchers demonstrate this through a simple mathematical question. They first posed a question about Oliver picking kiwis: as follows: Oliver picked 44 on Friday.

Biotechnology company Chai Discovery has received nearly $30 million in investment from Thrive Capital and OpenAI within six months of its founding. The company focuses on the application of biotechnology in drug development, developing innovative technologies that can predict biochemical molecular structures and reprogram molecular interactions, aiming to transform biology into 'engineering.' Chai Discovery's goal is to accelerate and optimize the drug development process by enhancing drug molecule design through technologies such as artificial intelligence and machine learning.

MIT research reveals that AI, particularly chatbots, can implant "false memories" in the human brain. Experiments show that AI can lead volunteers to develop persistent incorrect memories through questioning. Individuals interested in but unfamiliar with AI are more susceptible to this influence. This finding poses challenges for AI applications in law, healthcare, and education, emphasizing the need to ensure AI acts as a tool for protecting and....

KAN2.0, an upgraded version of the neural network architecture, aims to deepen the integration of AI and classical physics. Researchers can customize the model to incorporate expertise for discovering key concepts in physical systems, such as the Lagrangian. The new framework enhances the model's expressiveness, practicality, and interpretability through three main features: MultKAN, kanpiler, and Tree Converter, addressing compatibility issues between AI and science. KAN2.0 excels in feature recognition, revealing module structures, and discovering symbolic formulas, particularly in fields that are difficult to represent with symbolic equations, such as chemistry.

New research reveals the potential of large language models (LLMs) to significantly enhance performance through search capabilities, particularly in Python code generation tasks. The Llama3.1 model, with only 800 million parameters, performs comparably to GPT-4o after 100 search iterations. This finding resonates with Rich Sutton's view in 2019 that the integration of general methods and search and learning strategies is crucial in the face of increased computational power. Researchers found that increasing the number of resampling during the reasoning phase can significantly improve the model's performance.

The AI painter struggles to generate specific scenes like 'Ice Cola in a teacup,' highlighting the issue of text-image misalignment. Zhao Juntu, a PhD student at Shanghai Jiao Tong University, and his team discovered that even the most advanced AI models have difficulty accurately understanding and realizing complex concepts described in text, such as the difference between a transparent glass cup and a traditional teacup. To address this issue, they proposed a Mixture of Concept Experts (MoCE) method, utilizing large language models to help AI grasp hidden concepts, thereby enabling more precise control over text.

In collaboration with the International Olympic Committee and the Senegal National Olympic Committee, Intel has developed an AI platform that can monitor and evaluate athletes' performances using only smartphone cameras, identifying future Olympic talent. This platform analyzes video clips using computer vision to generate performance statistics, providing evaluators with the basis for talent discovery. In its initial application in Senegal, the platform analyzed over 1,000 children in six villages and identified 40 significantly talented athletes. The International Olympic Committee plans to launch a global project in 2025, leveraging AI technology to discover hidden talents.

Researchers from Stanford University and the University of California, Berkeley conducted an experiment to explore the consequences of retraining AI image generation models using images generated by the AI itself. The experiment utilized the Stable Diffusion (SD) model and selected high-quality facial images from the FFHQ dataset for automatic generation. The results showed that after retraining, the AI model produced highly distorted images. This distortion originated not only from the text prompts used for retraining but also persisted even when retraining was done solely with real images.

YouTube Music is back with a big announcement! Recently, they revealed they are testing two new features: one is an AI-generated conversational radio feature, and the other is a Shazam-like song identification tool. These two new features will undoubtedly make music discovery smarter and more personalized.AI Conversational Radio: Discover Music Through ChatThe AI conversational radio feature of YouTube Music is currently being tested with some Premium users in the United States. Users can create