Author's): Dr. José Crespo
Originally published in Towards Artificial Intelligence.
Though still haunted by the AI's deadly obsession with Euclidean Flatland
Yes, ChatGPT, Claude, Gemini, all of them. Magnificent fossils of a linguistic era that is now coming to an end. They decompose in public places, billions are still spent on polishing their coffins: bigger models, longer contexts, more hallucinations per watt.
The article discusses the decline of traditional language models (LLMs) such as ChatGPT, positioning them as obsolete as the industry turns to new AI architectures focused on world modeling rather than text generation. The author introduces a new model called JEPA, suggesting that it can provide true intelligence by predicting reality rather than merely processing language. However, criticism persists that JEPA and similar models still use flawed methodologies rooted in outdated mathematical principles, which, if not addressed, could lead to further problems in the development of artificial intelligence. Ultimately, the author argues for a necessary paradigm shift in the understanding and architecture of AI systems.
Read the entire blog for free on Medium.
Published via Towards AI
