Chinese AI start-up DeepSeek has developed a new method to enhance the reasoning abilities of large language models (LLMs) through self-principled critique tuning (SPCT), enabling better and faster responses than competitors. Their AI model, R1, claims to perform comparably to ChatGPT while being more cost-effective. Collaborating with Tsinghua University, DeepSeek's approach involves a real-time judging system that evaluates AI responses, leading to self-improvement. The company plans to release these models as open-source software, with speculation about a new R2 chatbot. Teljes cikk (Euronews.com)