How the world’s most powerful open source artificial intelligence model was created

Last Monday, about a dozen engineers and senior executives from data science and artificial intelligence companies data block Gather in a conference room connected via Zoom to find out if they succeeded in building a top-notch platform AI Language model.The team spent months and approximately $10 million training DBRX, a large language model The design is similar to the one on the back OpenAI’s ChatGPTBut they won’t know how powerful their creation is until the results of the final proficiency test come out.

Jonathan Frankle, chief neural network architect at Databricks and leader of the DBRX team, ultimately told the team, “We’ve surpassed it all.” The team responded with cheers, cheers, and applause. Frankle usually avoids caffeine, but after staying up all night writing up results, he took a few sips of his iced latte.

Databricks will release DBRX under an open source license, allowing others to build on its work. Data shared by Frankle showed that in about a dozen benchmarks that measure an AI model’s ability to answer general knowledge questions, perform reading comprehension, solve vexing logic puzzles, and produce high-quality code, DBRX outperformed others. all good Available open source models.

Artificial Intelligence Decision Makers: Jonathan Frankle, Naveen Rao, Ali Ghodsi and Hanlin Tang.Photography: Gabriella Hasband

it stands out Meta’s Alpaca 2 and Mistral blends, two of the most popular Open source artificial intelligence model When the scores came in, Databricks CEO Ali Ghodsi enthused, “Yeah! Wait, did we beat Elon’s stuff?” Frankle replied that they did beat the Grok AI model. Recently open sourced by Musk’s xAIadding, “If we get one mean tweet from him, I’ll consider that a success.”

To the team’s surprise, DBRX also came very close across multiple scores to GPT-4, OpenAI’s closed model that powers ChatGPT and is widely considered the pinnacle of machine intelligence. Frankel said with a smile.

building module

By going open source, DBRX Databricks adds more momentum to a movement challenging the secretive practices of the most prominent companies in the current generative AI craze. OpenAI and Google keep the code of their GPT-4 and Gemini large language models strictly confidential, but some competitors, Notable metadatahave released their model for others to use, arguing it will spur innovation by putting the technology in the hands of more researchers, entrepreneurs, startups and established businesses.

Databricks said it also wants to be public about the work involved in creating its open source model, while Meta has not disclosed some key details about creating its open source model. camel 2 modelThe company will publish a blog post detailing the work involved in creating the model and inviting WIRED to spend time speaking with Databricks engineers as they make critical decisions in the final stages of the multi-million dollar DBRX training process. Building leading AI models is complex and challenging, and the latest innovations in the field promise to reduce costs. This, combined with the availability of open source models like DBRX, shows that AI development isn’t slowing down anytime soon.

Ali Farhadi, CEO Allen Institute for Artificial Intelligence” said that there is an urgent need to improve the transparency of artificial intelligence model construction and training. The field has become increasingly mysterious in recent years as companies seek an advantage over their competitors. Opacity is particularly important when people worry about the possible risks of advanced artificial intelligence models. “I’m happy to see any efforts to open up,” he said. “I do believe that a large part of the market will move to an open model. We need more of these efforts.”

Source link

Related Articles

Leave a Reply

Your email address will not be published. Required fields are marked *

Back to top button