iKala
The article discusses the technological divergence between the US and China in AI development, highlighting DeepSeek's new DeepSeek-OCR model. This model proposes a method to efficiently compress conversation history into visual representations, retaining key information while minimizing memory load. The potential implications for long-term AI interactions and cost optimization in AI development are significant.
Mira Murati's Thinking Machines Lab has launched its first product, Tinker, aimed at facilitating AI model fine-tuning for developers. Tinker simplifies the process by managing infrastructure and providing a flexible API for deeper control. This platform signifies a shift to "Fine-tuning-as-a-Service," democratizing AI customization for various users beyond large tech companies.
AI researchers have achieved a groundbreaking advance in neural networks by developing a method that allows graph neural networks to solve shortest path problems across graphs of vastly different sizes. By aligning the neural network's learning process with the logical steps of classical algorithms and using a technique called "sparsity regularization," they've created AI systems […]
DeepSeek Shakes Up AI, Forcing Big Tech to Step Up DeepSeek has everyone on their toes. Just last week, Google launched Gemini 2.0 Flash, and now they've suddenly rolled out three new models, declaring the official dawn of the Gemini 2.0 era. Meanwhile, OpenAI, clearly spurred by DeepSeek, rushed to release o3-mini, followed quickly by […]
Is DeepSeek Truly Open Source? Debunking the Debate I see people starting to debate whether DeepSeek is truly open source, shouting that they've only released the model weights and explained training methods in papers, claiming this isn't real open source! Where's the training code? Where's the training dataset? Understanding the Core of Open-Source AI: The […]