According to reports, two people familiar with the matter said today that artificial intelligence startup OpenAI has stopped developing a new artificial intelligence model Arrakis. As early as the end of last year, after the chatbot ChatGPT caused a global sensation, OpenAI engineers began working on a new artificial intelligence model, internally codenamed "Arrakis".
According to the original design intention of OpenAI, Arrakis will have the same capabilities as GPT-4, but the deployment cost will be lower because part of its design adopts the so-called "decentralization" principle.
This means that only part of the neural network is used to process user input. In traditional "dense models", the entire neural network is active. Currently, Google’s PathAI project uses the “dispersion” principle.
But two people familiar with the matter said today that OpenAI canceled the Arrakis project in the middle of this year because the model was not operating as efficiently as expected. OpenAI's development team realized that Arrakis' performance was far inferior to GPT-4.
It's unclear why the "dispersion" principle worked so well in early testing but failed in later development.
The current large language model, that is, generative artificial intelligence, relies on huge computing power to run its complex mathematical models, and companies must purchase a large number of high-performance chips, which is very costly. As a result, more and more companies, including Amazon, Google, Alibaba, Meta and OpenAI, are designing their own specific artificial intelligence chips.
But analyst firm CCSInsight said last week that generative AI has been overhyped. As associated costs increase and calls for regulation grow, this emerging technology will be out of business by 2024.