This comprehensive survey by Yashar Deldjoo explores the advancements in recommender systems (RS) using generative models (Gen-RecSys). It covers key advancements in interaction-driven generative models, the use of large language models (LLMs) for natural language recommendation, and the integration of multimodal models for generating and processing images and videos. The survey highlights the need for evaluating the impact and harm of Gen-RecSys and identifies open challenges in the field. The core contributions of the survey include: 1. A broader scope than previous surveys, encompassing a wide array of generative models in RS. 2. Classification of models based on the type of data and modality they are used for. 3. In-depth exploration of deep generative model paradigms across multiple contexts and use cases. 4. Detailed evaluation of Gen-RecSys, including benchmarks, impact and harm evaluation, and conversational evaluation. 5. Discussion of several open research challenges and issues. The survey is structured into four main sections: 1. **Generative Models for Interaction-Driven Recommendation**: Discusses auto-encoding models, auto-regressive models, generative adversarial networks (GANs), diffusion models, and other generative models for user-item interactions. 2. **Large Language Models in Recommendation**: Explores the use of LLMs for dense retrieval, generative recommendation, retrieval-augmented recommendation, and conversational recommendation. 3. **Generative Multimodal Recommendation Systems**: Focuses on the challenges and approaches for multimodal recommendation, including contrastive and generative methods. 4. **Evaluating for Impact and Harm**: Reviews evaluation metrics for offline impact, online and longitudinal evaluations, and societal impact, emphasizing the complexity of evaluating Gen-RecSys. The survey concludes with conclusions and future directions, highlighting important challenges and opportunities in the field of Gen-RecSys.This comprehensive survey by Yashar Deldjoo explores the advancements in recommender systems (RS) using generative models (Gen-RecSys). It covers key advancements in interaction-driven generative models, the use of large language models (LLMs) for natural language recommendation, and the integration of multimodal models for generating and processing images and videos. The survey highlights the need for evaluating the impact and harm of Gen-RecSys and identifies open challenges in the field. The core contributions of the survey include: 1. A broader scope than previous surveys, encompassing a wide array of generative models in RS. 2. Classification of models based on the type of data and modality they are used for. 3. In-depth exploration of deep generative model paradigms across multiple contexts and use cases. 4. Detailed evaluation of Gen-RecSys, including benchmarks, impact and harm evaluation, and conversational evaluation. 5. Discussion of several open research challenges and issues. The survey is structured into four main sections: 1. **Generative Models for Interaction-Driven Recommendation**: Discusses auto-encoding models, auto-regressive models, generative adversarial networks (GANs), diffusion models, and other generative models for user-item interactions. 2. **Large Language Models in Recommendation**: Explores the use of LLMs for dense retrieval, generative recommendation, retrieval-augmented recommendation, and conversational recommendation. 3. **Generative Multimodal Recommendation Systems**: Focuses on the challenges and approaches for multimodal recommendation, including contrastive and generative methods. 4. **Evaluating for Impact and Harm**: Reviews evaluation metrics for offline impact, online and longitudinal evaluations, and societal impact, emphasizing the complexity of evaluating Gen-RecSys. The survey concludes with conclusions and future directions, highlighting important challenges and opportunities in the field of Gen-RecSys.