Search AI Products and News

Intel's former CEO Pat Gelsinger collaborates with Gloo to launch the Flourishing AI (FAI) benchmark, evaluating AI models' alignment with human values. Based on Harvard's prosperity research, it covers 6 dimensions including virtues, relationships, and happiness, adding a unique 'faith & spirituality' category. The benchmark aims to guide AI development toward human well-being.....

In the field of artificial intelligence, reward models are a critical component for aligning large language models (LLMs) with human feedback, but existing models face the issue of "reward hacking." These models often focus on superficial features, such as the length or format of responses, rather than identifying genuine quality metrics, such as factual accuracy and relevance. The root cause lies in standard training objectives failing to distinguish between spurious associations and true causal drivers present in the training data. This failure leads to fragile reward models (RMs), which generate misaligned policies.

While most delivery automation systems are still stuck at the sidewalk, Rivr wants to bring robots directly to your doorstep. The robotics startup, headquartered in Zurich, is pilot testing a four-wheeled robot that can climb stairs with US logistics company Veho in Austin, Texas. This robot aims to solve the last tough part of end-to-end delivery: the final 100 yards. Designed to complete the most complex and physically demanding leg of the journey between Veho's van and the customer's door, this robot is being put through its paces by Rivr founder and CEO.

On April 24th, Kunlun Wanwei announced the official open-sourcing of its multimodal reasoning model, Skywork-R1V2.0 (hereinafter referred to as R1V2.0). This upgraded version demonstrates significant improvements in both visual and text reasoning capabilities, particularly excelling in deep reasoning for challenging science problems in the College Entrance Examination and general task scenarios. It is considered the currently most balanced open-source multimodal model, equally adept at visual and text reasoning.

OpenAI recently released its latest AI model, GPT-4.1, claiming superior instruction following. However, independent tests suggest a decline in alignment, i.e., reliability, compared to its predecessor, GPT-4. OpenAI typically releases detailed technical reports including safety evaluations with new models, but hasn't done so this time, explaining that GPT-4.1 is not considered a 'cutting-edge' model.

Anthropic, an AI company, recently published a significant study analyzing the values expressed by its AI assistant, Claude, in real-world conversations. By deeply analyzing 700,000 anonymized conversations, the research team revealed 3,307 unique values demonstrated by Claude across various contexts, offering new insights into AI alignment and safety. This research aimed to assess whether Claude's behavior aligns with its design goals. The research team developed a novel evaluation method...

On March 6, 2025, a new language model called **Light-R1-32B** was officially launched. Built upon the **Qwen2.5-32B-Instruct** model, this mathematical problem-solving tool, specially trained for exceptional mathematical abilities, low training costs, and reproducibility, is a significant highlight in the field of artificial intelligence. The development team xAI stated that Light-R1-32B not only surpasses similar models in performance but also provides significant benefits for academic research and practical applications.

Zhou Hongyi discusses the challenges posed by AI hallucinations and their implications for cybersecurity. He emphasizes the need for a balanced approach, neither downplaying nor overstating the risks involved.

Recently, an AI system developed by Google DeepMind—AlphaGeometry2—has successfully surpassed the average level of gold medalists at the International Mathematics Olympiad (IMO), demonstrating outstanding performance in solving geometry problems. AlphaGeometry2 is an upgraded version of the AlphaGeometry system released by DeepMind last year. According to the latest research from the research team, this system can solve 84% of the geometry problems from the IMO over the past 25 years.

Generative language models face many challenges during the transition from training to practical applications. One major issue is how to achieve optimal performance during the inference stage. Current strategies, such as Reinforcement Learning from Human Feedback (RLHF), primarily focus on improving the model's win rate but often overlook decoding strategies during inference, such as Best-of-N sampling and controlled decoding. This gap between training objectives and actual usage may lead to inefficiencies and affect the quality and reliability of outputs. To address these issues, Google D

In the United States, despite the apparent benefits of recycling, the recycling rate has stagnated in recent years. To address this challenge, AMP Robotics has launched an AI system called AMP ONE to help waste processing plants sort garbage more efficiently. Recently, the company completed a $91 million Series D funding round to further advance the adoption of this technology. Image source note: Image generated by AI, image licensed from MidjourneyAMP Robotics.

The Alibaba International AI team has recently released a new reasoning model called Marco-o1, which focuses on solving open-ended questions, extending beyond traditional subjects with standard answers such as programming and mathematics. The research team is dedicated to exploring whether such models can be effectively adapted to fields that are difficult to quantify and lack clear rewards.

At the 2024 World Baidu Conference, Baidu founder Li Yanhong presented a series of eye-catching advancements in the field of AI. He pointed out that the most significant breakthrough in large model technology over the past two years has been the basic resolution of the 'illusion' problem, marking a qualitative leap in the accuracy of AI responses, transforming it from 'seriously spouting nonsense' to a reliable intelligent assistant. At the conference, Li Yanhong emphasized Baidu's latest retrieval-augmented generation technology, iRAG, which innovatively integrates the vast image resources accumulated by Baidu's search platform and advanced techniques.