Shedding light on…n°6 - ChatGPT and the rise of conversational AI models
ChatGPT has put conversational AI models in the spotlight. Capable of conversing with a human being, these technologies raise many questions about their numerous applications. This 6th issue of “Shedding light on…” invites us to take a technical dive into the heart of these devices in order to understand their main challenges and limits.
GPT-4, Stable Diffusion, GitHub Copilot: many generative models of artificial intelligence have emerged in recent months. Thanks to their performance and ergonomics, they are no longer reserved for experts only, but are now open to the public at large. Among them, the so-called “conversational” models have generated a lot of reactions. Capable of responding to requests formulated in natural language, they can carry on a conversation with a human being, whatever the level of technicality. Their popularity is already proven (ChatGPT has over 100 million monthly active users) and their applications are numerous.
How are these conversational models built? What place do they occupy in the Natural Language Processing (NLP) ecosystem? How are they innovative? This issue of “Shedding light on…” invites us to dive into the heart of the technology of these devices in order to understand their main challenges and limits. Combining several pre-existing technical bricks, these AI models are not so revolutionary compared to previous NLP models, but rather innovate in terms of accessibility. Their design raises many questions about privacy, intellectual property or openness of science: where GPT models are developed as real products, more respectful and open models could also find their place. Educating users would allow us to go beyond the hype around only a few models and truly democratize their capabilities.
This file may not be suitable for users of assistive technology, you can request an accessible format