Mistral just released a super powerful OCR engine

Hello pals! Here’s what you need to know about AI today:

👉 Mistral AI’s multi-lingual powerful OCR engine is here

👉 DuckDuckGo AI features exit beta

👉 ChatGPT is now able to edit code directly on MacOS

and many more!

📧 Did someone forward you this email? Subscribe here for free to get the latest AI news everyday!

Read time: 5 minutes

What’s going on: Mistral, a French leading AI company, has launched a new API called Mistral OCR for developers who work with PDF documents. This API converts PDFs into text files, which is important because large language models work best with raw text, and companies need clean data formats for AI processing. Unlike other OCR APIs, this one is multimodal, meaning it can identify and include illustrations and photos within the text by creating bounding boxes around them.

What does it mean: Instead of producing a single block of text, Mistral OCR formats the output in Markdown, a syntax commonly used by developers to add formatting elements like links and headers. This is particularly useful because LLMs rely on Markdown for training, and AI assistants use it to format outputs like bullet lists and bold text. According to Mistral's co-founder, Guillaume Lample, Mistral OCR helps companies convert complex documents into readable content, making it easier to adopt AI assistants for internal documentation.

More details: 

• Mistral OCR is accessible through Mistral's API platform or cloud partners like AWS, Azure, and Google Cloud Vertex. It can also be deployed on-premises for companies handling sensitive data.

• The company claims that its OCR performs better and faster than APIs from Google, Microsoft, and OpenAI, especially with complex documents containing mathematical expressions, advanced layouts, or non-English text.

• Mistral is using this technology in its own AI assistant, Le Chat, and anticipates that developers will use it with RAG systems for various applications, such as law firms sifting through large volumes of documents.

• Interested to get into the details and see the examples? Check out this page.

What’s going on: DuckDuckGo, a privacy-concerned search engine, is increasing its focus on generative AI, as its AI chat interface, duck.ai, exits beta. Duck AI allows users to access AI models from Anthropic, OpenAI, and Meta through a conversational interface, providing AI-generated answers to search queries instead of traditional search results. DuckDuckGo is expanding the models available on its AI chat service, including OpenAI's o3-mini, Meta's Llama 3.3, and Mistral's Small 3.

What does it mean: DuckDuckGo is increasing the frequency of AI-assisted answers in its conventional search engine, displaying generative AI text summaries above the usual links. Users can choose how often these AI-generated answers appear, maintaining control over the level of AI integration in their search results. The company emphasizes privacy, allowing users to access major GenAI tools without needing to sign up for an account with an AI provider. Chats on Duck AI are anonymized and not used for AI model training.

More details:

• DuckDuckGo offers a chat button to easily switch between its conventional search engine and the AI chat interface. They have expanded the sources for AI-assisted answers beyond Wikipedia and now include sources from across the web, providing links with more information.

• A "Recent Chats" feature stores conversations locally on the user's device, with the option to disable storage for increased privacy.

• DuckDuckGo emphasizes that its AI chats are not used for AI training and ensures that any temporarily stored chats are deleted within 30 days by its providers.

💻 ChatGPT now allows direct code editing on MacOS in developer tools like Xcode and VS Code for ChatGPT Plus, Pro, and Team subscribers; positioning it as a stronger competitor to AI coding tools like Cursor and GitHub Copilot.

🏭 Larry Page, the co-founder of Google, is developing a new AI startup named Dynatomics, which focuses on utilizing AI to optimize designs for product manufacturing.

💸 Turing, a platform providing engineers to contribute code to AI projects including building LLMs for OpenAI and others, raised $111 million doubling its valuation to $2.2 billion.

👨‍💻 Claude Code, Anthropic's AI coding tool, experienced a rocky launch due to a bug in its auto-update function that, when installed with root or superuser permissions, altered critical system file permissions, potentially rendering systems unstable or even unusable.

💰 Intangible AI, a no-code, AI-powered 3D creation tool for filmmakers and game designers raised $4 million to launch its web-based platform in June. The company is led by former Apple and Unity designer Charles Migos.

🔍 The US Department of Labor is investigating Scale AI, a data labeling startup valued at $13.8 billion, for potential violations of the Fair Labor Standards Act regarding unpaid wages, worker misclassification, and illegal retaliation.

📃 Anthropic submitted AI policy recommendations to the White House, including preserving the AI Safety Institute, developing national security evaluations for AI models, creating a government team to analyze AI security vulnerabilities, hardening AI chip export controls (particularly to China), and establishing a national target of building 50 additional gigawatts of power for AI data centers by 2027.

AI + Therapy

Genies - Lead Machine Learning Engineer, 3D Gen AI & Graphics

Roku - Senior Data Scientist, Ads Manager

Intuitive - Sr ML Software Engineer

XPENG - Staff Machine Learning Engineer – End-to-End Autonomous Driving

Thank you for staying with us like always! If you are not subscribed, subscribe here for free to get more of these emails in your inbox! Cheers!