Generative AI: Innovation and Efficiency in Building Technology Solutions

Luis Fernando Ortega, Technology Manager at Nuvu, presented a talk on "Generative AI: Innovation and Efficiency in Building Technology Solutions". He began by explaining key concepts such as language modeling, tokens, embeddings and large language models (LLMs), highlighting how these models, trained with vast amounts of data, overcome the limitations of recurrent neural networks (RNNs) through the use of transformers. These advances enable efficient parallelism and scalability with GPUs, significantly improving the models' processing power and context handling.

Ortega delved into the emerging skills of LLMs, illustrating with examples the importance of well-defined prompts for optimal results. He explained how to formulate accurate and contextual prompts, and the need to review and adjust the responses generated to ensure their accuracy and relevance. In addition, he addressed the challenges of inherent biases in training data and strategies to mitigate them, such as document retrieval, detailed refinement, and the use of diverse and representative data.

Finally, Ortega shared practical experiences in the implementation of generative AI, showing concrete applications such as news classification and test scoring. He also mentioned the importance of organizational technological maturity in areas such as hybrid cloud, security, DevOps, monitoring and data management. He concluded by inviting to explore the creation of applications that assist in various industries, and recommended resources to delve deeper into the development of LLMs for production environments.