_{Welcome back for the Latest AI Drops!}

Today’s Drops:

• Child Abuse Imagery Detected in Stable Diffusion’s AI Image Training Dataset

• AI-Driven Discovery of New MRSA Antibiotics at MIT

• UPS Uses AI To Mitigate Package Theft

• Trending AI Research Papers by Google & Alibaba Group

• Trending on X

• AI News “Anthropic is negotiating a significant funding round, Musk's AI 'Grok' earns surprising praise, the UK Supreme Court rules on AI's legal status as inventor, a Japanese startup innovates AI chip technology, and Google Ads announces major AI enhancements.”

• Trending GitHub Projects

• AI Tools

Read Time: 4.5 minutes

Child Abuse Imagery Detected in Stable Diffusion’s AI Image Training Dataset

A recent Stanford Internet Observatory report revealed the presence of child sexual abuse material (CSAM) in the LAION-5B dataset, which Stability AI uses for its Stable Diffusion model. The report found over 1,000 instances of illegal images in this dataset, primarily compiled from social media and adult websites. The dataset indexes links to these images along with alt text rather than storing the images directly.

This discovery raises severe concerns about AI image generation technologies' safety and ethical integrity. LAION, the organization overseeing the dataset, has implemented a zero-tolerance policy towards harmful content and temporarily withdrawn the datasets from public access. Stability AI claims to have used only a portion of LAION-5B and made efforts to ensure safety.

The Stanford report highlights the risks of training AI models on datasets containing inappropriate content. The report suggests that indirect exposure to CSAM can significantly influence AI behaviour. It also emphasizes the challenges in removing problematic content from AI models once trained on such datasets, leading to recommendations for discontinuing the distribution of models trained on LAION-5B, such as Stable Diffusion 1.5.

In response to these findings, US attorneys general are calling for a Congressional investigation into AI's role in child exploitation and for legislative action to prohibit AI-generated CSAM. This incident underscores the urgent need for stricter dataset curation and model training protocols to ensure AI technologies' responsible and safe development.

Source

AI-Driven Discovery of New MRSA Antibiotics at MIT

Scientists at MIT, led by James Collins, have made a remarkable breakthrough in antibiotic research using AI. For the first time in over 60 years, they've discovered a new class of antibiotics to combat the challenging methicillin-resistant Staphylococcus aureus (MRSA). This discovery, detailed in a study published in Nature, involved a deep learning model that analyzed approximately 39,000 compounds for their potential to fight MRSA while minimizing human cell toxicity.

This AI-driven approach led to the identification of two promising antibiotic candidates from a massive screening of 12 million compounds and provided insights into the AI's decision-making process. These candidates have shown effectiveness in reducing MRSA in mouse models, marking a significant advancement in the fight against antibiotic resistance and showcasing the transformative role of AI in medical research and drug development.

Source

UPS Uses AI To Mitigate Package Theft

UPS is leveraging artificial intelligence to combat the rising issue of package thefts through its innovative program, DeliveryDefense. This AI-driven system focuses on reducing theft risks by analyzing delivery addresses for potential vulnerabilities. Here's how it works:

1. AI Risk Analysis: The program uses AI to assess the likelihood of theft at a specific delivery address. It calculates a risk score based on past delivery experiences and other relevant factors.

2. Scoring System: Each address is given a score. A higher score indicates a lower risk of theft, while a lower score suggests a higher risk. This scoring helps in preemptively identifying addresses that are more prone to theft.

3. Proactive Measures for Risky Deliveries: For addresses that receive a low score (high risk), the program offers merchants the option to reroute the package. The package can be sent to a secure location like a UPS Store or another pickup location with the customer's agreement.

4. Statistical Impact: According to UPS Capital's president, Mark Robinson, about 2% of addresses are categorized as high-risk, accounting for approximately 30% of the total losses due to theft. This statistic underscores the significance of identifying and addressing these high-risk areas.

5. Real-World Application: Businesses dealing with high-value items, like Texas Precious Metals, find this system particularly useful. It helps them ensure the safe delivery of items like gold coins and silver bars, which are especially vulnerable to theft.

UPS's DeliveryDefense system is a prime example of how AI can be applied to solve logistical challenges and enhance the safety and reliability of package deliveries, particularly during peak times like the holiday season.

Source

AI Research Papers

1:VideoPoet: A Large Language Model For Zero-shot Video Generation

Research Paper by Google introduces VideoPoet, a large language model for zero-shot video generation that produces a range of large & smooth motions while preserving objects’ appearance over multiple seconds.

2:AnyDoor: Zero-shot Object-level Image Customization

Research paper by Alibaba Group introduces AnyDoor, a diffusion-based image generator with the power to teleport target objects to new scenes at user-specified locations in a harmonious way.

Trending on X

Welcome to the innovative world of GitHub projects!

1. Ucas-HaoranWei/Vary: Official code implementation for "Vary: Scaling Up the Vision Vocabulary of Large Vision Language Models"​​.

2. damo-vilab/i2vgen-xl: VGen is an open-source video synthesis codebase developed by the Tongyi Lab of Alibaba Group, featuring state-of-the-art video generative models.

3. danswer-ai/danswer: Danswer is a tool for asking questions in natural language and getting answers backed by private sources, with integration capabilities for platforms like Slack, GitHub, and Confluence​​.

4. QwenLM/Qwen: Official repository for Qwen (通义千问), a chat and pre-trained large language model proposed by Alibaba Cloud​​.

5. PaddlePaddle/PaddleDetection: An object detection toolkit based on PaddlePaddle, supporting object detection, instance segmentation, multiple object tracking, and real-time multi-person key point detection​​.

AI News

Anthropic is negotiating a $750 million funding round led by Menlo Ventures, potentially valuing the AI startup at over $15 billion, reflecting strong VC interest in major AI deals. Link

Elon Musk's AI chatbot 'Grok' unexpectedly receives positive reviews despite its rapid development and initial perception as a publicity stunt, challenging existing AI chatbots like ChatGPT.Link

The UK Supreme Court has ruled that AI cannot be recognized as an inventor under current law, denying Stephen Thaler's petition to name his AI system DABUS as the inventor of specific products. This decision aligns with similar rulings in the US, emphasizing the legal stance that inventors must be human or a company, and highlights ongoing debates about the legal status of AI-generated creations. Link

Japanese startup Preferred Networks is designing its own AI chips to address the global demand and bottlenecks in AI technology, emphasizing optimized energy consumption and enhanced computing power for AI tasks.Link

Google Ads confirms ongoing support with upcoming major AI enhancements, addressing concerns about the withdrawal of the free feature and customer service issues.Link

AI Tools

Pic Copilot is an AI powered e-commerce image tool

Linnk.AI is the key to rapid, insightful content navigation. Linnk.AI cuts through content chaos, offering clear, AI-powered insights for busy researchers and multilingual users.

LogoFast helps make beautiful logos with AI, fast & free

Snowball helps content creators grow on X by writing personalized posts and replies optimized for engagement with AI 

Shader helps create AI filters in seconds