Despite the fact that generative artificial intelligence (AI) models supported by centralized cloud infrastructure, such as ChatGPT, are now leading the pack in terms of overall performance, findings from a recent study indicate that open-source competitors are beginning to catch up.
Companies like Google and OpenAI, which are currently at the forefront of the generative artificial intelligence market, have taken a centralized approach to the construction of their infrastructure.
The centralized approach has effectively restricted public access to a variety of information, including the data sources utilized for the training model. There is a possibility that this will change, according to the research team at Cathy Wood’s ARK Invest, which suggests that open-source artificial intelligence models may outperform their centralized equivalents by the year 2024.
MMLU performance of open-source and private AI models. Source: AKR Invest
Open-source AI models have made significant development since 2022, as shown in the graph above. Some of these models have ended up performing better than private models. With well-known models such as ChatGPT-4 and Gemini Ultra, OpenAI, Google, and Alphabet, the parent company of Google, are the dominant players in the centralized artificial intelligence area.
A number of Chinese artificial intelligence models, as well as Meta (which was once known as Facebook), have chosen to implement an open-source methodology. Yi 34B, Falcon 180B, and Mixtral 8x7B emerged as some of the most prominent open-source artificial intelligence systems in 2023, with performance that was comparable to that of market leaders.
Jozef Soja, a researcher at ARK Invest, made the observation that Mixtral outperformed GPT 3.5 in terms of absolute log error of performance on Massive Multitask Language Understanding (MMLU) benchmarks. He also brought attention to the fact that GPT-4 is currently “just how far ahead of the pack.”
In addition, the underlying model of Meta, known as LLaMA, also demonstrated a notable improvement in its 2023 iterations. The importance of democratizing access to generative artificial intelligence is the driving force behind the demand for performance in open-source artificial intelligence models, according to AI engineer Brian Roemmele.
In both instances, artificial intelligence implemented a similar strategy when it came to warning consumers about investing advice and recommending that they seek the assistance of a professional financial adviser in order to receive more “personalized advice.”
Furthermore, both AI models delivered responses that highlighted similar topics of consideration, despite the fact that they offered different details in certain instances.