OpenAI, the company known for its cutting-edge research in artificial intelligence, recently stopped developing its new AI model called “Arrakis“. This model was intended to be a more efficient and economical version of the current GPT-4 model.
Development of Arrakis began last fall, and model training began this spring. But OpenAI developers quickly realized that the system was too slow. For a month they tried to customize the model, but then management closed the project. The reasons for the failure have not yet been established: the sparsity principle worked well in early tests, but failed to manifest itself in a larger model.
Despite the failure, OpenAI will be able to use the developments of Arrakis and the Gobi multimodal model in other projects. Microsoft also had high hopes for Arrakis: the software giant has now switched to developing less expensive language models, as the high cost of maintaining generative AI began to threaten its business.
The failure of the Arrakis project is a reminder that in the world of artificial intelligence, not all ideas are destined to succeed. However, even failures can become valuable experience that will be used for further research and development.
