The Daily Newsletter for Intellectually Curious Readers

If you're frustrated by one-sided reporting, our 5-minute newsletter is the missing piece. We sift through 100+ sources to bring you comprehensive, unbiased news—free from political agendas. Stay informed with factual coverage on the topics that matter.

Join for free today!

Today:

• Altera's Breakthrough in Digital Humans

• HeyGen’s Avatar 3.0

• Google Rolls Out AI Security Feature

• Inflection AI Introduces Agentic AI Models

• TikTok’s AI Tool Rivals Meta, Google

• Squarespace Unveils AI-Powered Design Tools

Altera's Breakthrough in Digital Humans

Altera, founded by Dr. Robert Yang, aims to create "digital humans" that collaborate with people using AI inspired by human brain functions. By leveraging OpenAI's GPT-4, Altera addresses the challenge of data degradation, allowing AI agents to interact autonomously for extended periods. 

Their first product, AI agents that play Minecraft like friends, demonstrates how these digital humans can form meaningful, long-lasting interactions. Altera’s vision extends beyond gaming, aiming to build AI agents for various applications, from problem-solving coworkers to entire virtual societies, pushing the boundaries of human and AI collaboration.

HeyGen’s Avatar 3.0

HeyGen has launched Avatar 3.0, allowing users to create AI-powered clones of themselves with customizable poses, outfits, and camera angles. To make an avatar, visit HeyGen’s site, click "Avatars," and follow the steps to create a video of yourself. Once your avatar is ready, create a video by selecting "Create Video," choosing your avatar, and adding a script. 

You can also add scenes to showcase different looks. After submitting, the video takes a few minutes to generate, depending on its length, and you're done.

Google Rolls Out AI Security Feature

Google is rolling out an AI-powered Theft Detection Lock for Android phones, which senses sudden movement, such as snatching, and immediately locks the screen. This feature uses the phone's gyroscope and accelerometer to detect theft-like motion and secure your device. 

Some users may already have it on their phones, such as Xiaomi 14T Pro or Pixel devices. Additional security features include harder factory resets, private spaces for sensitive apps, offline locks, and remote locking via a number. The update is being distributed through Google Play services, with some features exclusive to Android 15.

Inflection AI Introduces Agentic AI Models

Inflection AI has introduced Inflection 3.0 and a commercial API aimed at addressing the issue of uniformity in AI model outputs caused by Reinforcement Learning with Human Feedback (RLHF). By incorporating employee feedback into its RLHF process, Inflection tailors AI models to align with a company’s culture, focusing on emotional intelligence (EQ). 

The company’s approach allows enterprises to fine-tune AI on-premise for added security. Inflection also emphasizes “agentic AI” (Action Quotient), enabling its models to not only empathize but also take meaningful actions, such as automating tasks, setting it apart from other generative AI tools.

TikTok’s AI Tool Rivals Meta, Google

TikTok has launched Smart+, its AI-powered advertising tool designed to compete with Meta’s Advantage+ and Google’s Performance Max. Smart+ automates ad creation, targeting, and optimization, offering marketers the flexibility to either manage campaigns themselves or let the AI take full control. This modular approach allows users to selectively leverage the tool’s capabilities. 

Early test results show promising outcomes, with brands like Ray-Ban achieving improved return on investment and reduced acquisition costs. TikTok aims to attract marketers seeking faster, simplified campaign management while competing for a larger share of ad dollars in the performance advertising space.

Squarespace Unveils AI-Powered Design Tools

Squarespace has introduced Design Intelligence, a new website builder that integrates generative AI tools, aiming to enhance the design process rather than overwhelm users with generic content. Chief Product Officer Paul Gubbay emphasized the importance of balancing AI automation with customization, allowing users to retain creative control while benefiting from AI-generated designs, text, and images. 

Squarespace focuses on curating content using proprietary prompts, ensuring the results align with the brand’s design standards. Gubbay stressed that AI is an opportunity to enhance design efficiency, not replace human creativity, making website creation faster without sacrificing quality or uniqueness.

🧠RESEARCH

Addition is All You Need for Energy-efficient Language Models

Researchers propose a new algorithm, L-Mul, which replaces energy-intensive floating-point multiplications with simpler integer additions. This method cuts energy costs by up to 95% while maintaining high precision in neural network tasks. Testing shows L-Mul delivers comparable or better accuracy than traditional methods across various benchmarks, making it highly efficient.

DepthPro Demo with 3D Visualization

Depth Pro is a cutting-edge model for zero-shot metric monocular depth estimation, producing sharp and detailed depth maps without needing metadata like camera specifics. It generates high-resolution (2.25 MP) depth maps in just 0.3 seconds on standard GPUs. The model uses an efficient multi-scale vision transformer and leverages real and synthetic data for precise boundary accuracy and metric depth. It also excels at focal length estimation from a single image, providing fast and highly accurate depth predictions.

DreamWaltz-G: Expressive 3D Gaussian Avatars from Skeleton-Guided 2D Diffusion

"DreamWaltz-G" introduces a novel framework for generating 3D avatars from text using 2D diffusion models guided by skeletons. By integrating 3D human templates and hybrid 3D Gaussian representations, the model enhances avatar quality, addressing issues like multiple faces or extra limbs. It enables real-time rendering, expressive animations, and supports applications such as video reenactment and scene composition. The method surpasses previous approaches in both visual and animation quality.

Tutor CoPilot: A Human-AI Approach for Scaling Real-Time Expertise

"Tutor CoPilot" presents a Human-AI system that helps tutors by providing expert-like guidance in real-time. Tested in a study with 900 tutors and 1,800 students, the system improved student mastery by 4 percentage points, especially benefiting lower-rated tutors. Tutor CoPilot encourages high-quality teaching strategies and reduces costs, making expertise more accessible in education, particularly for underserved communities. 

CANVAS: Commonsense-Aware Navigation System for Intuitive Human-Robot Interaction

"CANVAS" introduces a framework for commonsense-aware robot navigation using visual and linguistic instructions. CANVAS learns human-like navigation through imitation learning, trained on the COMMAND dataset with over 48 hours of annotated results. It outperforms traditional systems, excelling in noisy environments and achieving strong Sim2Real transfer success. This approach enables robots to navigate intuitively by interpreting abstract human commands, showing significant improvements in unseen environments and real-world applications.

🛠️TOP TOOLS

Base AI - Build serverless autonomous AI agents with memory

Twig - Drafts responses to your customer queries referencing help articles, resolved tickets and support channels.

Selfletter - Break complex goals into simple tasks with AI

Costgpt - Simplifies software project planning by providing comprehensive roadmaps that include detailed estimates like costs, features, sitemaps, and milestones.

BuyScout - Your AI Copilot for Online Shopping

📲SOCIAL MEDIA

🗞️MORE NEWS

• KoBold Metals raised $491 million to support its AI-driven search for minerals crucial for clean energy, including copper, lithium, and cobalt. With over 60 projects, it recently discovered a significant copper deposit in Zambia. This funding will help KoBold develop the resource and expand exploration efforts.

• Google is rolling out its AI-powered Gmail Q&A feature for iOS. This assistant helps users quickly summarize emails, find information, and manage inbox tasks. It's available to Google One AI Premium and specific Workspace subscribers.

• Apple is moving away from its annual product upgrade cycle, opting for more flexible releases. This shift allows for staggered product launches, reducing delays and easing the burden on engineers while still maintaining innovation and revenue outside of the holiday season.

• Google Photos has started rolling out the AI-powered "Ask Photos" feature to select users in the US. This tool allows users to search their photo libraries using natural language prompts, such as finding Halloween costumes or specific events. Initially available through a waitlist, it replaces the "Search" tab in the app. Results are more accurate when asking about people or locations, with varying success for events or activities.

• Researchers at Cleveland Clinic, in collaboration with IBM, are using AI to discover non-opioid pain relief options. Their deep-learning tool, LISA-CPI, identifies FDA-approved drugs and gut metabolites that can target pain receptors without addiction risks. Studies are underway to validate the findings.

• The US Department of Health and Human Services (HHS) has funded a project using AI to accelerate antibiotic discovery. The TARGET project aims to identify 15 promising antibiotics by using generative AI and deep learning to screen and assess molecular compounds for antibacterial activity. Led by Phare Bio and MIT-Harvard collaborators, the project seeks to combat rising antibiotic resistance with a budget of up to $27 million.