Alibaba's latest technique lowers AI training costs by almost 90%
What's the story
Chinese tech giant Alibaba Group has unveiled a revolutionary technique that could cut the cost of training artificial intelligence (AI) models for search engines by a huge margin.
The new approach, dubbed ZeroSearch, has achieved an impressive 88% reduction in costs.
The development comes as companies across the globe are working to improve their AI capabilities while keeping development costs low.
Innovation
A cost-effective AI training solution
Alibaba's researchers have developed ZeroSearch to enhance the search capabilities of AI models through simulations.
With this innovative approach, there's no need to interact with actual search engines, as they detailed in a recently published research paper.
By allowing AI models to generate quality content in response to queries, ZeroSearch removes high costs associated with routing queries through commercial search engines.
Accessibility
A game-changer for resource-limited teams
ZeroSearch turns a reference model into a search engine capable of training other AI systems to respond to queries.
This development drastically minimizes dependence on expensive external search infrastructure, making AI search training more affordable.
It especially helps smaller teams with limited resources, who can now train their own AI models without paying prohibitive costs involved in conventional methods.
Cost analysis
Cost comparison: ZeroSearch vs traditional methods
A practical example highlights how cost-effective ZeroSearch really is.
Sending queries to Google 64,000 times through an application programming interface (API) would usually cost around $586.70.
However, using a 14-billion-parameter AI model to generate responses for training would only cost $70.80.
That's an impressive cost reduction of nearly 88%, which further highlights the potential benefits of Alibaba's innovative AI training approach.