Welcome to the AI Daily section! This is your daily guide to exploring the world of artificial intelligence. Each day, we bring you the hottest topics in the AI field, focusing on developers, helping you understand technological trends and innovative AI product applications.
Fresh AI Products Click to Learn More: https://top.aibase.com/
1. Challenging NVIDIA! AMD Unveils Its Most Powerful AI Chip, the Ryzen AI 300 Series
AMD showcased its latest AI chip lineup at the Computex technology conference, featuring the Zen5 architecture Ryzen 99950X processor and the Ryzen AI 300 series APU, challenging NVIDIA's dominance in the AI field.
AiBase Highlights:
🚀 NVIDIA and AMD showcase their latest technological achievements at the Computex technology conference, with AMD unveiling the Zen5 architecture Ryzen 99950X processor and the Ryzen AI 300 series APU.
💥 Lisa Su emphasized that the Zen5 architecture Ryzen CPU has a wider CPU engine instruction window, supports full AVX512 throughput, and doubles AI performance.
🔥 AMD's Ryzen AI 300 series APU uses the XDNA AI NPU, with a computing power of up to 50 TOPS, surpassing the performance standards of other competitors.
2. Suno Set to Launch New Feature! Hum a Few Notes and It Can Create a Song for You
Suno recently announced an exciting new feature that can generate complete songs from humming sounds, showcasing the limitless possibilities of AI in music creation. Suno's innovative approach injects new vitality into music creation, allowing users to create music from everyday sounds, opening up new possibilities for music creation. We look forward to Suno's future innovations that will amaze us.
AiBase Highlights:
🎵 Humming to Song Creation: Suno's new feature allows users to create a complete song from a short hum, with music and vocals blending naturally.
🎶 Transforming Everyday Sounds into Music: Suno's new feature can turn any sound into a music piece, showcasing the powerful creative potential of the technology.
🎤 Inspiring Music Creation: Suno's innovative approach injects new vitality into the music creation field, opening up new possibilities for music creation.
More details: https://top.aibase.com/tool/suno-ai
3. Kunlun Wanwei Announces the Open Source of the 200 Billion Parameter Sparse Large Model Skywork-MoE
Kunlun Wanwei has open-sourced the landmark sparse large language model Skywork-MoE, which boasts strong performance and significantly reduced inference costs, providing an effective solution for large-scale dense LLMs.
AiBase Highlights:
🌟 Open Source and Free Commercial Use: The Skywork-MoE model weights and technical reports are fully open-sourced and free for commercial use, promoting the development of the AI field.
💡 Reduced Inference Costs: Skywork-MoE significantly reduces inference costs while maintaining performance, addressing the challenges of large-scale data processing.
🚀 Technological Innovation and Performance Advantages: Skywork-MoE is the first open-source trillion-parameter MoE large model that supports inference on a single 4090 server, with strong performance and a large number of parameters.
More details: https://top.aibase.com/tool/skywork-moe
4. Adobe Releases VideoGigaGAN Super-Resolution Video Model
Adobe, in collaboration with researchers, has introduced VideoGigaGAN, a super-resolution video model that balances frame rate consistency and rich details. This model addresses the issues of temporal consistency and detail richness in super-resolution video models, bringing significant breakthroughs to the video processing field.
AiBase Highlights:
⭐ VideoGigaGAN is developed based on the GigaGAN model, adding temporal convolution, self-attention layers, and optical flow guidance modules to solve the issues of temporal consistency and detail richness in super-resolution video models.
⭐ VideoGigaGAN uses temporal convolution to capture temporal dependencies between video frames, self-attention layers to extract spatial details and texture information, and optical flow guidance modules to maintain spatial consistency, generating clear super-resolution videos.
⭐ VideoGigaGAN features video super-resolution, temporal consistency, rich detail processing, anti-aliasing, and is suitable for various video processing scenarios.
More details: https://top.aibase.com/tool/videogigagan
5. Stanford University AI Research Team Accused of Plagiarizing Tsinghua-Affiliated Model
This article reports on the controversy surrounding the Stanford University AI research team's Llama3-V open-source model, which has been accused of plagiarizing the open-source model "Mini Cannon" MiniCPM-Llama3-V2.5 developed by the Tsinghua-affiliated startup FaceWall Intelligence. After the incident was exposed, the two main authors from the Stanford team apologized to the FaceWall Intelligence team and the public, and promised to remove all Llama3-V models.
AiBase Highlights:
🔍 The Stanford University AI research team's Llama3-V model has been accused of plagiarizing the MiniCPM-Llama3-V2.5 model developed by the Tsinghua-affiliated startup FaceWall Intelligence.
🚨 Netizens discovered that the structure and code of the Llama3-V model are highly similar to the "Mini Cannon" model, sparking widespread attention and discussion.
🔗 The FaceWall Intelligence team confirmed the plagiarism, and the two main authors from the Stanford team publicly apologized on social platforms and promised to remove all Llama3-V models.
More details: https://top.aibase.com/tool/minicpm-llama3-v-2-5
6. Multimodal Model Evolves Further, Now Learns to Play Poker and Calculate "12 Points" from Images
This article introduces the new reinforcement learning framework RL4VLM proposed by the research teams from UC Berkeley and other institutions, which successfully improves the performance of multimodal large models in decision-making tasks. The model, through reinforcement learning fine-tuning, has learned to play poker and calculate "12 points" from images, surpassing GPT-4v. The research team consists of several heavyweight figures, and the results have been open-sourced on GitHub.
AiBase Highlights:
🧠 The new reinforcement learning framework RL4VLM successfully improves the decision-making ability of multimodal large models, surpassing GPT-4v.
🌟 The research team includes heavyweight figures such as Turing Award winner LeCun.
💡 RL4VLM uses reinforcement learning fine-tuning, directly using environmental reward information, giving multimodal models the ability to make autonomous decisions.
Paper address: https://arxiv.org/abs/2405.10292
Project address: https://top.aibase.com/tool/rl4vlm
7. OpenAI Spin-off Company's AI Model Enables Robots to Think and Learn Like Humans
This article introduces the AI model developed by OpenAI spin-off company Covariant, which enables robots to think and learn like humans. The model combines reasoning and physical capabilities, featuring multimodal input, autonomous task execution, feedback and interaction, adaptability, and other characteristics, representing a significant advancement in robot learning and automation technology.
AiBase Highlights:
🤖 The Covariant AI system combines reasoning skills and physical dexterity, developing the RFM-1 model to handle multiple input types, enabling robots to comprehensively understand task requirements.
🧠 Robots can autonomously execute tasks, adapt to the environment based on feedback and interaction requests, without relying on specific task code, simplifying the programming process.
🔗 Covariant's AI system gives robots the ability to visually recognize, think, act, and learn, improving the flexibility and efficiency of automation.
8. Bitcoin Miners Invest Millions in AI Companies, Seeking Billions in Returns
Bitcoin miner Core Scientific has partnered with cloud company CoreWeave in a $3.5 billion deal to expand its AI business, capitalizing on the growing demand in the AI field. This move will bring substantial revenue, driving the transformation of miner companies to cope with the challenges posed by the halving of Bitcoin.
AiBase Highlights:
⭐ Bitcoin miner Core Scientific has partnered with CoreWeave in a $3.5 billion deal to expand its AI business.
⭐ The AI field has huge demand, providing substantial revenue and driving the transformation of miner companies.
⭐ Bitcoin miners are seeking diversified revenue in the AI market to cope with the challenges of halving.
9. IBM Introduces Efficient LLM Benchmarking Method, Reducing Computational Costs by 99%
IBM Research has introduced an innovative LLM benchmarking method that significantly reduces the time and money costs required to evaluate LLMs through miniaturized benchmarking, drawing attention from the AI community and有望推动人工智能模型评估领域的快速发展。
AiBase Highlights:
⭐️ Innovative LLM benchmarking method reduces computational costs by 99%.
⭐️ Efficient method utilizes miniaturized benchmarking, reducing evaluation time and money costs for LLMs.
⭐️ Gains attention from the AI community and, if widely adopted, could drive development in the field of AI model evaluation.
10. McKinsey Global Survey: Generative AI Adoption Begins to Generate Value
The widespread adoption of AI is transforming how organizations operate, but it also brings some negative impacts. McKinsey's survey shows that AI is reducing costs and increasing revenue in areas like marketing and sales, but inaccuracies and safety remain concerns. High-performing organizations face more challenges with GenAI adoption but succeed through best practices.
AiBase Highlights:
⭐️ 65% of organizations regularly use AI, with GenAI widely applied across multiple fields, leading to cost reduction and revenue growth.
⭐️ 44% of respondents have experienced negative impacts from GenAI use, including inaccuracies, cybersecurity risks, and intellectual property infringements.
⭐️ High-performing organizations face more challenges with GenAI adoption but succeed by increasing risk awareness, establishing clear processes, and developing employee skills.
