Hey folks, hope you had a good weekend! On our side we had snow and it was beautiful.

On the AI side here is our input : we're super excited by the xAI announcement because this team is moving really, really fast and in the right directions (don't forget that the team got created only 1 year ago). They're also really available on X and are waiting for your feedback, as this post shows. We think this is the right time to test Grok, so please let us know your most funny interactions with it!

We also spent time to improve the visual quality of the newsletter, please give feedback!

Top AI news

1. Grok AI Chatbot Now Free for All Users
xAI has announced that its Grok-2 AI chatbot is now free for all users on the X platform. This updated version is three times faster and offers improved accuracy, better instruction-following, and enhanced multilingual support. Premium and Premium+ subscribers will enjoy higher usage limits and early access to new features. Grok now includes web search, citations, and the Aurora image generator, enabling real-time responses and photorealistic image creation. Additionally, xAI has launched a public beta API, allowing developers to incorporate Grok's capabilities into their applications. This move aims to democratize AI access and foster innovation across various sectors.

2. AI Models That Understand Your World
AI world models are a groundbreaking approach inspired by human mental representations, enabling systems to predict, simulate, and plan for real-world scenarios. These models enhance robotics by anticipating outcomes, making machines more adaptive and effective in complex environments. In video generation, they produce realistic sequences by accurately modeling object interactions and movement. Companies like World Labs, founded by Fei-Fei Li, are investing heavily in this technology, applying it across diverse fields such as gaming, simulation, and beyond. This innovation marks a significant step towards creating smarter, more autonomous AI systems capable of revolutionizing industries and transforming how machines interact with the world.

3. Overreliance on Synthetic Data Threatens AI Stability
AI models trained predominantly on synthetic data risk performance degradation, known as "model collapse." This occurs as errors are propagated through recursive training on AI-generated outputs, diminishing accuracy and reliability. Integrating human-sourced data introduces necessary diversity and authenticity, enhancing model robustness and preventing decline. Balancing synthetic and human data is crucial to ensure AI systems remain effective and trustworthy across various applications.

Tech Bonus - Microsoft's Phi-4: 14B Parameter AI Model
Microsoft has unveiled Phi-4, a 14-billion parameter small language model (SLM) that excels in complex reasoning tasks, especially in mathematics. Phi-4 outperforms larger models by leveraging high-quality synthetic datasets and post-training innovations. Currently available on Azure AI Foundry under a Microsoft Research License Agreement, Phi-4 will be accessible on Hugging Face next week. Microsoft emphasizes responsible AI development, providing robust tools to measure, mitigate, and manage AI risks throughout the development lifecycle.

Tweet of the Day

Interesting conversation between Musk and Google CEO after quantum Google announcement.

Startup Review of the day

Devin AI: The Future of Autonomous Software Development
Discover how Devin AI by Cognition Labs automates coding, debugging, and deployment, boosting engineering productivity and enabling teams to focus on innovation.

Stay Connected

Feel free to contact us with any feedback or suggestions—we’d love to hear from you !