The past year has seen the continued rapid advancement of large language models, yet training them has become increasingly complex. Even companies with vast financial and computational resources are encountering significant limitations that slow progress in enhancing these models’ capabilities.
One notable example is OpenAI’s development of its new Orion model, which has faced several challenges. While Orion has shown some improvements in language tasks, its overall progress—particularly in programming—has been modest compared to the leap from GPT-3 to GPT-4. Testing has revealed that the model struggles to handle coding tasks as efficiently as its predecessor, raising serious concerns within the development team.
The whole article is available here (in Slovene language only).
Download the entire article:
Author: Jenz Perš
