Cognitive Asia

[In Photos] Spotlight Falls on 'Robots + Cultural Tourism' at Industry Expo in Shanghai

Yicai Global | English | News | April 3, 2025 | UndeterminedUndetermined

The Tourism Plus Shanghai exhibition commenced on March 30, 2025, at the Shanghai World Expo Exhibition & Convention Center, showcasing the innovative integration of 'robots + cultural tourism.' The five-day event spans 700,000 square meters and features over 6,000 exhibitors, highlighting advancements across the entire cultural tourism supply chain, including dining, accommodation, and entertainment.

Attendees are invited to engage with various robotic technologies, such as the robot dog from Unitree Robotics and the dancing, coffee-making androids from Qingbao Engine Robot. Visitors can also experience Ossur Intelligence's hiking exoskeleton, inspired by the popular film The Wandering Earth, and interact with Unitree's humanoid robot, which gained fame for its performance on New Year's Eve television.

Additionally, the expo offers immersive experiences like the extended reality exhibition "Dunhuang: Realm of a Thousand Colors," marking its Shanghai debut and further enhancing the cultural tourism experience at the event.

台積電赴美設廠 重電股受惠 亞力盤中大漲逾7%

TSMC's plant in the US benefits from heavy electronics stocks; Alex surges more than 7% during trading

United Daily News | Local Language | News | April 3, 2025 | UndeterminedTech Development/Adoption

No Body Summary

생성형 AI, 지능 평가했더니…1등은 'IQ 130', 누구?

Generative AI, Intelligence Evaluation... Who is ranked first with an 'IQ of 130'?

ZD Net Korea | Local Language | News | April 3, 2025 | UndeterminedUndetermined

Google's Gemini 2.5 Pro has achieved the highest intelligence ranking among AI models, scoring an IQ of 130 on the Mensa Norway test. This score reflects a level of intelligence that is deemed "very good" for humans, indicating that generative AI is nearing human cognitive abilities. In comparison, OpenAI's o1 Pro and o1 models scored 120 and 125, respectively, while Anthropic's Claude 3.7 Sonnet Extended scored 107 points on the same test.

The results also highlighted OpenAI's dominance in the space, with seven of the top ten AI models belonging to the company. This includes models such as o1 Pro and GPT-4.5 preview. Anthropic followed closely with their Claude 3.7 Sonnet Extended and Claude 3.7 (Vision) models, securing the 3rd and 9th positions. In contrast, models from DeepSeek, xAI, and Meta ranked lower on the list.

An analysis of domain-specific performance revealed that AI models show significant variability in effectiveness between text-based and vision-based tasks. For example, GPT-4o demonstrated contrasting scores in text and vision processing. Meanwhile, OpenAI's o1 Pro displayed balanced performance across both domains. Trends in AI development are shifting towards models with enhanced reasoning capabilities, as seen with Google's Gemini 2.0 Flash Thinking and xAI's Grok-3 Think, suggesting an increase in AI's problem-solving capabilities.

Lastly, while IQ tests can measure certain cognitive skills, they may not fully represent an AI's overall practical capabilities in real-world scenarios. Various factors like context understanding and user intent are crucial in determining the best model for specific applications. Users can access these AI models via web interfaces or APIs, with major services available from OpenAI, Anthropic, and Google.