Search AI Products and News

Large Language Models (LLMs) have achieved significant progress in complex reasoning tasks by combining task prompts with large-scale reinforcement learning (RL), as demonstrated by models like Deepseek-R1-Zero, which directly apply reinforcement learning to base models, showcasing strong reasoning capabilities. However, this success is difficult to replicate across different base model families, especially within the Llama series. This raises a core question: what factors lead to inconsistent performance of different base models during reinforcement learning? How does reinforcement learning perform in

X-ORIGIN-AI (Xuan Yuan Technology), a consumer-grade AI robot company, recently announced it has completed a Pre-A round of financing totaling nearly 100 million yuan. The round was led by Oriental Fortune Capital, with participation from Kingtop Capital and Lenovo Star, and Renchen Capital served as the financial advisor. This funding accelerates X-ORIGIN-AI's development in AI hardware and affective interaction, showcasing its breakthroughs in the development of emotionally intelligent AI products. X-ORIGIN-AI is dedicated to breaking the limitations of traditional AI tools and pushing human-computer interaction from a 'tool-based' to an 'emotionally intelligent' approach.

On March 31, Qianxun Intelligence (Spirit AI), a leading company in the embodied AI field, announced the completion of a Pre-A round of funding totaling 528 million RMB. This round was led by Prosperity7 Ventures (P7), an affiliate of Aramco Ventures, with participation from several renowned investors including China Merchants Venture Capital, GF Securities, Jingya Capital, Orient Capital, and Huacontrol Investment.

Researchers at the University of California, Berkeley have recently released their latest research achievement – the TULIP (Towards Unified Language-Image Pretraining) model. This model aims to enhance the performance of vision-language pre-training, particularly in visually-centric tasks requiring high-fidelity understanding, overcoming the limitations of existing contrastive learning models (such as CLIP). TULIP integrates innovative techniques such as generative data augmentation, enhanced contrastive learning, and reconstruction regularization.

The automotive market is experiencing another wave of intelligent upgrades! Leapmotor, a leading domestic new energy vehicle brand, today announced that its new model, the Leapmotor B10, will officially begin pre-sales on March 10th. This new car, upon its unveiling, immediately ignited market anticipation with its hard-core configuration of "150,000 RMB class first-ever laser radar urban intelligent driving" and cutting-edge technology of "built-in DeepSeek and Tongyi Qianwen dual AI models." It has been hailed by the industry as the 'Intelligent Driving King' of the 150,000 RMB automotive market! According to the latest pre-release information from Leapmotor, the Leapmotor B10...

According to CNBC, Perplexity, the AI search engine developer, is raising a $50 million seed and pre-seed investment fund. The fund is largely capitalized by limited partners, with Perplexity contributing a portion of its previous funding as cornerstone capital. In December, Perplexity successfully raised $500 million, valuing the company at $9 billion. This investment fund is managed by two...

Last Thursday, domestic tech company Shanji launched its first AI camera glasses, marking an important milestone in the domestic AI eyewear field. The glasses are priced at 1499 Yuan, while the first batch of 50,000 co-creation units was sold at a discounted price of 999 Yuan, reportedly even lower than similar products from Xiaomi. Additionally, Shanji introduced a special promotion where users can receive a full refund if they check in 200 days within 300 days (not necessarily consecutively), which undoubtedly increases the product's appeal.