![](\"https:\/\/s3.cointelegraph.com\/uploads\/2023-12\/9ab164d7-5898-4d25-a3f1-a2a6ac079e09.jpeg\")
<\/p>\nWhile generative artificial intelligence (AI) models backed by centralized cloud infrastructure \u2014 such as ChatGPT \u2014 currently lead on overall performance, new research shows that open-source competitors are catching up.<\/p>\n
The current market leaders of generative AI, such as Google and OpenAI, took a centralized approach to building their infrastructure \u2014 effectively limiting public access to various information, including the data sources used for the training model. <\/p>\n
This could change, the research team at Cathy Wood\u2019s ARK Invest claims, suggesting the possibility of open-source AI models outperforming their centralized counterparts by 2024.<\/p>\n
<\/p>\n
The above graph shows the progress made by open-source AI models since 2022, a few of which ended up performing better than private models. OpenAI, Google and its parent company, Alphabet, dominate the centralized AI space with popular models like ChatGPT-4 and Gemini Ultra. On the other hand, Meta (formerly Facebook), Mistral and a few Chinese AI models opted for an open-source approach. <\/p>\n
In 2023, Yi 34B, Falcon 180B and Mixtral 8x7B emerged as some of the top open-source AI that showcased comparable performance to market leaders. ARK Invest researcher Jozef Soja noted that Mixtral beat GPT 3.5 on absolute log error of performance on Massive Multitask Language Understanding (MMLU) benchmarking while highlighting \u201cjust how far ahead of the pack GPT-4 is currently\u201d. <\/p>\n
Additionally, Meta\u2019s foundational model, LLaMA, also recorded a significant improvement in its 2023 iterations. The need for performance in open-source AI models stems from the importance of democratizing access to generative AI, said AI engineer Brian Roemmele.<\/p>\n