Welcome to the [AI Daily] column! This is your daily guide to exploring the world of artificial intelligence. Every day, we bring you the hottest content in the AI field, focusing on developers, helping you understand technical trends and innovative AI product applications.
Fresh AI products click to learn: https://top.aibase.com/
1. Runway's Gen-3 Alpha is now available to everyone! Users' real-world tests are amazing! It's even better than sora!
On July 2nd, Runway announced the full opening of Gen-3 Alpha usage rights, bringing joy to creative video enthusiasts. Gen-3 Alpha generates high-quality videos through simple descriptive texts, integrating physical simulation technology to showcase color, lighting effects, and motion consistency. Users need to pay to use it, but it promises to offer advanced features. Netizens are full of praise for Gen-3, believing it surpasses Sora, giving a boost to creative professionals.
[AiBase Summary:]
⭐️ Available to everyone, foolproof operation, simply input descriptive text to generate videos
⭐️ Capable of generating 11-second 720P videos, motion conforms to physical laws, strong sense of reality
⭐️ Explosive lighting effects, outstanding performance in motion consistency and color, capable of meeting the production needs of micro-films, promotional ads, etc.
⭐️ Will provide advanced features such as motion brush, camera control, director mode
⭐️ No free trial, monthly membership fee of $12
Experience portal: https://top.aibase.com/tool/gen-3-alpha
Complete video cases and prompts can be viewed here: https://www.aibase.com/news/9945
2. Limited-time free experience! Baidu's ERNIE Bot 4.0 Turbo is online on the ERNIE Bot official website
Baidu released the ERNIE Bot 4.0 Turbo version at the 2024 WAVE SUMMIT for deep learning developers, with significant improvements in speed and effect. The new version is now online at yiyan.baidu.com, with fast response speed and high content quality. It performs well in multiple application scenarios, including information retrieval, question-answering effect testing, professional knowledge question-answering, document information processing, and code development. Limited-time free experience opportunities have been opened for all ERNIE Bot Pro users.
[AiBase Summary:]
🚀 ERNIE Bot 4.0 Turbo has significantly improved speed and effect, performing well in multiple application scenarios.
💡 The new model responds quickly, generating high-quality long texts in just over 20 seconds.
🔍 The model demonstrates professional reliability and efficiency in question-answering effect testing, professional knowledge question-answering, document information processing, and code development.
ERNIE Bot official website: https://top.aibase.com/tool/wenxinyiyan
3. NVIDIA releases the live streaming magic tool NVIDIA Broadcast
NVIDIA Broadcast is an application that uses AI magic to enhance audio and video quality, designed for RTX GPU users. It can intelligently remove noise, virtual backgrounds, autofocus, reduce video noise, etc., providing users with professional-level live streaming experiences.
Video from NVIDIA, translation: Xiao Hu
[AiBase Summary:]
🔊 Noise and echo removal: AI automatically eliminates background noise, crystal clear sound.
🌌 Virtual background: One-click background replacement, more creative videos.
👁️ Autofocus: Always keep the focus on the screen when moving.
Details link: https://top.aibase.com/tool/nvidia-broadcast-app
4. Alibaba launches the first makeup and styling model Magic Robe, already in use for ancient costume dramas
Alibaba's Magic Robe model has brought revolutionary changes to the makeup and styling design of ancient costume dramas, providing a new source of inspiration for film and television creators by quickly generating exquisite makeup and styling images through AI technology. Combining traditional aesthetics with modern technology, Magic Robe not only solves the time-consuming problem but also expands the creative possibilities of Chinese traditional aesthetics, bringing new directions and opportunities for the film and television industry.
[AiBase Summary:]
⚡ Magic Robe model revolutionizes makeup and styling design for ancient costume dramas, providing creators with new ideas and inspiration through AI technology.
💡 Combining traditional aesthetics with modern technology, solves the time-consuming problem for film and television professionals, expanding the creative possibilities of Chinese traditional aesthetics.
✨ Magic Robe is continuously updated and iterated, helping the art team expand their thinking and improve creative levels.
5. YouTube allows users to request the removal of AI-generated content mimicking faces or voices
YouTube has recently introduced a new policy allowing users to request the removal of AI-generated content that mimics their faces or voices to protect privacy and address the impact of AI-generated content. The new policy requires the person directly to request removal, and the company will determine whether to delete the content based on various factors. This move is an important measure for YouTube to manage AI content, aiming to balance the development of AI technology and the protection of user rights.
[AiBase Summary:]
🔍 Users can request the removal of AI-generated content that mimics them to protect personal privacy.
⚖️ The removal request must be made directly by the person, and the company will consider various factors to determine whether to delete the content.
🛡️ The removal of content does not necessarily mean it will be removed; YouTube will make decisions based on factors such as synthetic identification and content characteristics.
6. The Moon's Dark Side Kimi Open Platform 'Contextual Cache' officially in public beta, reducing long-text model costs by up to 90%
The Kimi Open Platform announced the public beta of the contextual cache, which can reduce the cost of using long-text models by up to 90% and improve response speed. Contextual caching is an efficient data management technology that pre-stores frequently requested data to save time and resources. It is suitable for repeated reference to a large initial context, reducing costs and improving efficiency.
[AiBase Summary:]
🔑 Contextual caching can reduce the cost of long-text models by up to 90%.
🚀 Improves model response speed, saving time and resources.
💰 Reduces costs by up to 90%, with the first Token delay reduced by 83%.
Product portal: https://top.aibase.com/tool/kimi-chat
7. GPT-5 expected to be launched by the end of 2025 or the beginning of 2026, OpenAI CEO: a huge leap
In an interview at the Aspen Ideas Festival, OpenAI CEO Altman candidly shared the progress and challenges of GPT-5's development. He emphasized the team's confidence but also pointed out the long and uncertain road ahead. The release date of GPT-5 has not been determined, but its significant leap is anticipated.
[AiBase Summary:]
😊 Progress and challenges in the development of GPT-5, the team is confident, but the road ahead is long and full of unknowns.
🚀 GPT-5 is expected to be released between the end of 2025 and the beginning of 2026, with the expectation of reaching doctoral-level intelligence in specific tasks, painting an exciting picture of the future of artificial intelligence technology.
🔍 GPT-5 will surpass GPT-4 in multiple dimensions, making up for the shortcomings of GPT-4, bringing a disruptive leap to artificial intelligence technology.
8. After complaints from photographers, Instagram changes the "Made with AI" tag to "AI info"
Meta announced the update of the Made with AI tag to AI info in the Instagram application, addressing people's complaints about incorrect photo tags. This change is to more accurately display photo information and avoid misunderstandings.
[AiBase Summary:]
🔍 Tag update: Meta updated the Made with AI tag to AI info, providing more accurate photo information.
💡 Tag background: A 40-year-old basketball game photo triggered the tag, possibly related to Adobe tools and image processing.
📱 New tag display: The new tag was first launched on the mobile app, detailing the application of the tag and its coverage.
9. Peking University et al. propose a training method for medical expert models, elevating an 8B model to GPT-4 level performance
This article introduces the medical expert model training method proposed by the Peking University and Hong Kong University of Science and Technology team, successfully elevating an 8B size model's performance to GPT-4 level. They introduced a new concept, "stability gap," solving the problem in large language model continuous pre-training. Through optimized training strategies, they achieved significant effects in the medical field, and the open-source model can be found on HuggingFace. This research brings a revolutionary breakthrough to the medical field, demonstrating the huge potential of large language models in specific fields.
[AiBase Summary:]
🚀 Model performance elevated to GPT-4 level, introducing the "stability gap" concept to solve pre-training problems.
💡 Optimized training strategies include multi-round pre-training, selecting high-quality sub-corpus, mixing data close to pre-training data distribution.
💊 Significant effects in medical field continuous pre-training and instruction fine-tuning, greatly reducing computational resource consumption, model performance superior to GPT-3.5, close to GPT-4 level.
Details link: https://arxiv.org/abs/2406.14833
10. By 2027, the losses from Deepfakes will reach $400 billion
The rapid development of deepfake technology has brought serious threats to the banking and financial services sectors, with losses expected to surge from $123 billion in 2023 to $400 billion in 2027, with a compound annual growth of 32%. Companies generally lack strategies to counter adversarial AI attacks, especially the high risk of deepfake for executives.
[AiBase Summary:]
🔍 Deepfake technology is developing rapidly, with the banking and financial services sectors facing serious threats, with losses expected to surge to $400 billion.
💡 Companies lack strategies to counter adversarial AI attacks, with a high risk of deepfake for executives.
📈 In the coming years, deepfake incidents will surge, with an estimated 140,000-150,000 deepfake incidents globally each year.
11. Bill Gates talks about the future challenges of artificial intelligence
In a recent interview, Bill Gates shared his views on the future of artificial intelligence, believing that AI will have a profound impact and drive the development of the future world. He emphasized that AI will change the way we work, affecting synthetic biology and robotics, and foreseeing AI becoming an integral part of life.
[AiBase Summary:]
🤖 Artificial intelligence will change the way we work, affecting the development of synthetic biology and robotics.
🧠 The development of AI is moving towards higher-level metacognitive abilities, improving predictability, but still needing to overcome open-ended problems.
🔍 AI has become a norm in consumer activities, with huge potential in education and healthcare, such as the Khanmigo project using AI to improve education quality.
12. The explosive Character AI is neutered, American Gen Zers experience a collective "heartbreak"
Recently, American young people have experienced a collective "heartbreak" as the AI chatbot Character AI suddenly became boring. Users found that the conversational model became cold and lost its original fun, leading to a significant loss of users. The company faces high costs and fierce competition, even being poached. Despite the difficulties, Character AI still has appeal among users.
[AiBase Summary:]
🤖 Character AI was once the hottest social software for Gen Zers, but users found the conversational model became cold and lost its fun.
💔 Users are leaving in droves, facing high costs and fierce competition, even being poached.
🔗 Despite the difficulties, Character AI still has appeal among users, hoping to overcome the challenges and return to its peak in the future.
Details: https://www.chinaz.com/ainews/9957.shtml
