The new AI model from OpenAI has proven to be less advanced than initially expected.

According to data from the publication Bloomberg, Orion apparently falls short in capabilities compared to its predecessor — the GPT-4 chatbot. A similar report by The Information indicated that some researchers at OpenAI believe there have been no improvements in certain areas, such as programming.

Moreover, OpenAI is not the only company facing challenges in AI development. The next iteration of Google's Gemini model has also failed to meet specialists' expectations, while the release timeline for the much-publicized Claude 3.5 Opus from Anthropic has been left "up in the air."

These industry-wide difficulties may signal that the paradigm of improving AI models through so-called scaling is hitting a wall, foreshadowing potential economic issues in the future if AI model development remains costly and significant leaps in performance towards creating general or strong artificial intelligence do not occur.

"The AI bubble is slowly bursting," stated Margaret Mitchell, the chief scientist for ethics at the AI startup Hugging Face. She noted that different approaches to training AI may be necessary to approach human-level intelligence and versatility.