New OCR from Mistral AI; Explore QwQ-32B Model on Qwen Chat & HuggingChat
New AI Models and Products
QwQ-32B Model Release: The QwQ-32B AI model has been made available on multiple platforms, including Qwen chat and HuggingChat. The model offers various functionalities such as image creation, coding, and planning. It is also available unquantized and for free on HuggingChat.
- QwQ 32B is available on Qwen chat now
- QwQ-32B is now available on HuggingChat, unquantized and for free!
Mistral OCR Model: Mistral AI has released a new OCR model designed to extract text from images and PDFs. The model is part of Mistral AI's suite of AI tools and is available as an advanced OCR API.
AI Research and Development
Meta's Latent Tokens for LLM Reasoning: Meta AI researchers have developed a method to improve the reasoning capabilities of Large Language Models (LLMs) by compressing text into latent tokens using a Vector Quantized Variational Autoencoder (VQ-VAE).
Perplexity AI's New Feature: Perplexity AI has added a new feature called 'Reasoning with Claude 3.7 Sonnet' to its platform, available to Pro users. This update aims to make customizing AI model settings more intuitive and faster.
AI Benchmarking and Comparisons
QwQ-32B vs. Qwen2 5-Coder 32B-Instruct: A benchmark comparison between 'qwq-32b' and 'Qwen2 5-Coder 32B-Instruct' in terms of coding tasks shows that 'qwq-32b' achieved 21% correctly completed tasks with a cost of around $0.8, while 'Qwen2 5-Coder 32B-Instruct' completed 8% with a cost of around $0.3.
AI Policy and Regulation
Anthropic's Warning to the White House: Anthropic has issued a warning to the White House regarding the R1 model, recommending that the U.S. government be equipped to evaluate future AI models for security risks.
AI Tools and Services
ChatGPT for macOS Code Editing: OpenAI has released a new feature for ChatGPT on macOS that allows it to edit code directly in IDEs. Users have reported mixed experiences with the feature.