Databricks, a $38 billion San Francisco-based startup, released open-source code on Friday that it says could be used by companies to create their own chatbots similar to OpenAI’s ChatGPT.
The code is an artificial intelligence model, an algorithm that is trained on sets of data and can then learn new data to perform a variety of tasks.
Databricks CEO Ali Qudsi said the post aims to show a viable alternative to training a type of AI model called a large language model with massive computing power and resources.
Read also : Are you looking for a smart phone? To check mobile locator
Large language model supporting OpenAI’s ChatGPT viral chatbot. OpenAI, valued at $29 billion, trains its AI models with a huge pool of data on a supercomputer from investor Microsoft Corp. OpenAI CEO Sam Altman said IT costs were “horrendous”.
OpenAI charges companies for access to its models for their own applications and expects it to hit $1 billion in sales by 2024.
The Databricks effort comes with caveats. Qudsi told Reuters that while the open-source chatbot has demonstrated impressive abilities in tasks such as writing blog posts, the company has not released official benchmark tests to show the bot matches performance. from ChatGPT.
Databricks sells cloud-based data mining and analysis software to enterprises and said last year it exceeded $1 billion in annual revenue. Databricks wants companies to train their AI models using their software. Qudsi said the company’s researchers took a two-year-old model available for free and trained it on a small amount of data for three hours on a single computer that anyone with a credit card could rent. “The future will be for everyone to have their own model, and they can actually train it, and they can improve it,” Qudsi said. This way they don’t have to give their data to anyone else either. ”
Databricks’ move comes as startups raise millions of dollars in venture capital investments to train their AI models, while big tech companies like Google and Alphabet’s Meta Platforms rush to downsize and scale. cost of AI models while improving their accuracy.
“I think eventually you’re going to make these models smaller and smaller and smaller, and they’ll be open source,” Qudsi said. “Everyone will have it. »