Databricks, a data science and AI company, has recently released DBRX, an open-source large language model that has outperformed all other open-source models. The model was developed over several months with an investment of about $10 million.
DBRX is similar in design to OpenAI’s ChatGPT and has shown superior performance across various benchmarks. These include general knowledge questions, reading comprehension, logical puzzles, and high-quality code generation. DBRX has even surpassed Meta’s Llama 2 and Musk’s xAI’s Grok AI model.
Interestingly, DBRX’s scores were also close to GPT-4, OpenAI’s closed model that powers ChatGPT and is considered the pinnacle of machine intelligence.
By open-sourcing DBRX, Databricks aims to spur innovation by making the technology accessible to more researchers, entrepreneurs, startups, and businesses. This move challenges the secretive approach of prominent companies in the generative AI boom, such as OpenAI and Google, who keep their code closely held.
Databricks plans to release more details about the creation of DBRX, which is a level of transparency that has not been seen from other companies like Meta regarding their Llama 2 model. The company has also hinted at expanding the availability of DBRX to more devices by the end of 2024.
read more > www.wired.com