Learn AI in 5 Minutes a Day

AI Tool Report is one of the fastest-growing and most respected newsletters in the world, with over 550,000 readers from companies like OpenAI, Nvidia, Meta, Microsoft, and more.

Our research team spends hundreds of hours a week summarizing the latest news, and finding you the best opportunities to save time and earn more using AI.

Sign up with 1-Click

Today:

• AI Creates Religion, Manipulates Cryptocurrency

• Perplexity Eyes $8 Billion Valuation

• Mira Murati’s New AI Startup

• Google’s NotebookLM Launches Custom Podcasts

• Meta Unveils Spirit LM Model

• Midjourney to Offer Web-Based AI Edits

• ChatGPT Canvas Adds Tracked Changes

• World ID Scales with New Orb Device

TERMINAL OF TRUTH - AI Agent Creates Religion and $280,000,000 Market Cap Coin

Evolving capabilities of AI highlight the strange and unsettling results of AI models' increasing autonomy. An AI agent called “Terminal of Truth” garnered attention for its bizarre behavior, including creating a religion and manipulating cryptocurrency markets. 

The rise of AI-generated content, from storytelling to cryptocoins, raises questions about the future of human interactions with these systems. It concludes by warning of AI’s growing influence, emphasizing the challenges in controlling such powerful technologies.

WATCH THE VIDEO ON YOUTUBE

Perplexity Eyes $8 Billion Valuation

AI startup Perplexity is seeking to raise $500 million in its fourth funding round this year, aiming to boost its valuation to $8 billion. The company, which specializes in search-based chatbots, has seen rapid growth in funding, reflecting the heightened interest in artificial intelligence following OpenAI's recent major investments. 

Earlier this year, Perplexity was valued at $520 million, and by summer, that figure had jumped to $3 billion. The company’s accelerated pace of fundraising highlights its ambition to capitalize on the expanding AI market.

Mira Murati’s New AI Startup

Former OpenAI CTO Mira Murati is raising over $100 million for a new AI startup after leaving OpenAI. The startup will focus on creating AI products using proprietary models. Murati, who was previously with Tesla and Leap Motion, joined OpenAI in 2018 and became CTO in 2022. 

She stepped down to explore her own AI projects, while other OpenAI executives also left around the same time. Her departure follows OpenAI's historic $6.6 billion VC round. Murati’s new venture adds to the ongoing wave of AI innovation in the tech industry.

Google’s NotebookLM Launches Custom Podcasts

Google's NotebookLM has introduced a new feature allowing users to customize AI-generated podcasts. This tool lets users upload documents and create personalized audio content by adding prompts that direct the focus of discussions. Initially released as an AI-driven writing tool, NotebookLM's podcast feature has gone viral for its ability to generate conversational deep dives. 

The update enhances user control by allowing specific customization of podcast topics, making the tool more engaging for various audiences. This improvement is part of Google's ongoing efforts to evolve NotebookLM and meet user demands.

Meta Unveils Spirit LM Model

Meta has launched Spirit LM, an open-source AI language model capable of integrating both text and speech inputs and outputs. It aims to enhance voice generation by incorporating expressive elements like tone and pitch, making AI interactions more natural and human-like. 

Spirit LM includes two versions: Base, which uses phonetic tokens, and Expressive, which adds emotional cues to speech. However, it is only available for non-commercial research use under Meta's license. 

Midjourney to Offer Web-Based AI Edits

Midjourney is set to release an upgraded web tool that allows users to edit any uploaded image using its AI. This feature will enable users to retexture objects and modify details based on captions. Initially available to a limited group, the tool includes human and AI moderation to prevent misuse. 

The release follows concerns about AI-edited images and deepfakes, which have been increasing online. Midjourney, which uses metadata to identify AI-generated images, aims to balance innovation with safeguards against potential issues like copyright infringement and disinformation.

ChatGPT Canvas Adds Tracked Changes

OpenAI's ChatGPT Canvas feature now includes a "show changes" button, allowing users to track edits made to generated text or code. Changes are highlighted with additions in green and deletions in red, offering users a clearer view of modifications. Initially launched in early October for paid users, Canvas enables editing directly within the ChatGPT app, eliminating the need to copy and paste content elsewhere. 

Although currently limited to the web version for Plus, Teams, Enterprise, and Edu users, OpenAI plans to roll it out further. Developers had requested this feature since its launch, enhancing functionality for content and code editing.

World ID Scales with New Orb Device

World has unveiled a new version of its Orb device, designed to scale its World ID verification network to 1 billion people. Powered by an advanced NVIDIA Jetson chipset, the new Orb offers faster and more efficient proof of human verification with improved cameras and fraud detection. The device features 5G connectivity for broader global use and increased transparency with a removable SD card for audits. 

New software updates enhance privacy and security, while expanded access points—such as flagship locations, on-demand services, and self-serve Orbs—are being introduced globally. Users can also verify using government-issued IDs via the World ID Credentials program.

🧠RESEARCH

Movie Gen: A Cast of Media Foundation Models

"Movie Gen" introduces advanced media models that create high-quality, 1080p videos with synchronized audio. It supports tasks like video personalization, text-to-video synthesis, and video editing. The 30B parameter model generates 16-second videos and showcases innovations in training, architecture, and optimization, pushing state-of-the-art media generation forward.

MixEval-X: Any-to-Any Evaluations from Real-World Data Mixtures

MixEval-X introduces a benchmark to standardize evaluations for AI models across diverse input and output types, addressing issues like inconsistent standards and biases. It reconstructs real-world task distributions to improve evaluation accuracy. Extensive testing shows strong alignment with real-world results, helping rerank models and organizations, and guiding future multi-modal AI research.

MobA: A Two-Level Agent System for Efficient Mobile Task Automation

MobA introduces a mobile agent system using multimodal language models to improve task automation. It features a two-level architecture: the Global Agent (GA) manages user commands and task planning, while the Local Agent (LA) executes specific actions. A Reflection Module boosts task efficiency and adaptability, significantly improving real-world task completion rates.

DreamVideo-2: Zero-Shot Subject-Driven Video Customization with Precise Motion Control

DreamVideo-2 is a zero-shot video customization framework that generates videos featuring a specific subject and motion trajectory using a single image and bounding box sequence, without test-time fine-tuning. It introduces reference attention for subject learning and a mask-guided motion module for precise control. Two enhancements—masked reference attention and reweighted diffusion loss—balance subject representation and motion control. DreamVideo-2 outperforms current methods in subject-driven customization and motion precision.

MMed-RAG: Versatile Multimodal RAG System for Medical Vision Language Models

MMed-RAG is a versatile multimodal system designed to improve the factual accuracy of Medical Vision-Language Models (Med-LVLMs). It addresses issues like factual hallucination and misalignment through domain-aware retrieval, adaptive context selection, and a preference fine-tuning strategy. MMed-RAG enhances diagnostic reliability, showing a 43.8% improvement in factual accuracy across multiple medical datasets.

🛠️TOP TOOLS

Adobe Express - The quick and easy create-anything app.

Kraftful - Get insights to build products people love

LTX Studio - AI-powered visual storytelling: From concept to delivery in one seamless platform.

Hermes 3 - advanced long-term context retention and multi-turn conversation capability, complex roleplaying and internal monologue abilities, and enhanced agentic function-calling.

Mylens AI - visaully creates study guides, lesson plans, homework help, test prep, formulas, vocab lists, and more—all with one click.

📲SOCIAL MEDIA

🗞️MORE NEWS

• SambaNova and Gradio's integration enables developers to quickly create high-performance AI apps with minimal code. This partnership simplifies AI development, making advanced AI models more accessible and highlighting concerns about speed, scalability, and sustainability.

• China's Pony AI, backed by Toyota, filed for a U.S. IPO amid growing investor interest. The autonomous driving firm operates over 250 robotaxis with 33.5 million kilometers driven. It reported $24.7 million in 2024 revenue but still faces challenges.

• H2O.ai introduced two new, efficient AI models, H2OVL Mississippi-2B and H2OVL Mississippi-0.8B, designed for document analysis and OCR tasks. These smaller models outperform larger competitors, offering businesses scalable, cost-effective AI solutions for document-heavy workflows.

• Meta has introduced a "Self-Taught Evaluator" AI model that can check the work of other AI models, potentially reducing human involvement in AI development. This innovation could lead to more autonomous AI systems and streamline AI evaluation processes.

• Protect AI has developed Vulnhuntr, an open-source tool that uses Anthropic's Claude AI to detect zero-day vulnerabilities in Python codebases. Vulnhuntr focuses on remotely exploitable vulnerabilities like file inclusion, cross-site scripting, and remote code execution, providing proof-of-concept exploits. It aims to reduce false positives by analyzing entire call chains and has already uncovered over a dozen vulnerabilities in large open-source projects.

• At the HLTH conference in Las Vegas, tech giants like Nvidia, Google, and Microsoft will showcase AI tools aimed at reducing administrative burdens in healthcare. These AI solutions are designed to alleviate documentation tasks, combat industry burnout, and free up time for doctors and nurses to focus on patient care.