Preeti
Lamini AI has developed a new technique called Lamini Memory Tuning that significantly improves the accuracy and reduces hallucinations in large language models (LLMs). This method involves tuning expert adapters with precise facts to embed them within the model, resulting in 95% accuracy and a reduction of hallucinations from 50% to 5%. The approach addresses the challenge of maintaining factual accuracy while preserving the generalization capabilities of LLMs.
By optimizing for zero error on specific facts, Lamini Memory Tuning enables near-perfect recall without compromising generalization. This has been successfully applied in critical applications, achieving 95% accuracy compared to 50% with traditional methods. The technique outperforms other approaches like Prompting and Retrieval-Augmented Generation by combining information retrieval with AI to eliminate incorrect responses effectively.
The innovative method involves creating memory experts tuned to recall specific facts with high fidelity, dynamically selected during inference. This results in a model that can scale to many parameters while maintaining low inference costs. Lamini Memory Tuning opens up new possibilities for LLMs in various industries, promising higher accuracy, lower costs, and faster development cycles.
As Lamini AI continues to refine this technology, the potential for fully automated, highly accurate AI-driven solutions grows. The implementation of Lamini Memory Tuning represents a new frontier in LLM development, with the promise of broader adoption and deployment in diverse applications.
Source link
Source link: https://www.marktechpost.com/2024/06/17/lamini-ais-memory-tuning-achieves-95-accuracy-and-reduces-hallucinations-by-90-in-large-language-models/?amp
GIPHY App Key not set. Please check settings