DeepSeek-R1 is a powerful reasoning model aimed at creating specialized AI systems for specific domains. The first-generation model, DeepSeek-R1-Zero, uses large-scale reinforcement learning (RL) without supervised fine-tuning, demonstrating strong reasoning capabilities but facing challenges like readability and language mixing [1]. To improve these aspects, DeepSeek-R1 introduces multi-stage training and cold-start data before RL, enhancing performance in reasoning tasks to levels comparable with leading models like OpenAI’s GPT-3. DeepSeek-R1, along with its predecessor DeepSeek-R1-Zero and six distilled models, is open-sourced to support the development of tailored AI solutions. This initiative provides a solid foundation for building efficient, domain-specific AI models for a range of

Ху Ц. (науч. рук. Балакшин П.В.) Building an AI-powered intelligent campus assistant with deepseek-r1 // Сборник тезисов докладов конгресса молодых ученых. Электронное издание. – СПб: Университет ИТМО, [2025]. URL: https://kmu.itmo.ru/digests/article/15363