Luna delivers highly accurate, low-latency results at near-free cost, enabling businesses to deploy trusted AI in production.
Galileo, the leader of development Generative AI We announced the release of Galileo Luna, the first suite of evaluation foundation models (EFMs) designed to transform how enterprises conduct generative AI evaluations. This novel approach is faster, more cost-effective, and more accurate than existing evaluation methods such as askGPT and human “vibe checks.” Galileo Luna enables enterprises to bring reliable AI solutions to market faster and at production scale.
read: Persistent Accelerates Digital Engineering with AI-Powered SASVA Platform
“For genAI to be widely adopted, it's important that companies can assess hundreds of thousands of AI responses in real time for hallucinations, toxicity, security risks, and more,” he said. Vikram ChatterjeeMike, co-founder and CEO of Galileo, said: “As we talked to our customers, we learned that existing approaches like human ratings and LLM-based ratings were too costly and time-consuming, so we set out to solve that. With Galileo Luna, we're setting a new benchmark for speed, accuracy, and cost-effectiveness. Luna can evaluate millions of responses per month for 97% less costly, 11x faster, and 18% more accurate than ratings using OpenAI GPT3.5.”
Luna: A breakthrough in AI evaluation technology
At the core of Luna's innovation is the creation of EFMs, the first models built specifically for generative AI assessment. Each of these models is fine-tuned to solve a specific assessment task, such as hallucinations, contextual quality, data leakage, and malicious prompt detection. By creating purpose-built miniature EFMs, Luna is able to conduct assessments with unprecedented accuracy, speed, and cost-effectiveness.
read: AI in Content Creation: Top 25 AI Tools
Galileo Luna's main innovations and features:
- Evaluation accuracy: Luna's EFM outperforms all existing evaluation models, including Galileo's proprietary Chainpoll, and leads the industry in detecting hallucinations, prompt injection, PII and more, performing up to 20% better than traditional methods.
- Ultra-low cost of operation: It has proven to be 30 times cheaper than traditional methods such as OpenAI's GPT 3.5.
- Millisecond speed: Designed for real-time applications, evaluation is completed in milliseconds, which is essential for real-time applications such as chatbots and AI monitoring systems.
- No ground truth needed: Unlike other evaluation methods that rely on extensive and costly test sets, Luna does not require ground truth data, enabling faster deployment and scalability.
- Unmatched customizabilityEach Luna model can be rapidly fine-tuned to specific customer needs, providing a customized solution that delivers greater than 95% accuracy in critical applications.
“Evaluation is absolutely essential to delivering safe, reliable, production-grade AI products,” he said. Alex Klug“Until now, existing assessment methods, such as human assessment or using LLMs as assessors, have been extremely costly and time-consuming. With Luna, Galileo is overcoming the biggest assessment hurdles for enterprise teams: cost, latency and accuracy. This is a game changer for the industry,” said Galileo, head of data science and AI products at HP.
Powering production generative AI applications.
At launch, Luna is already integrated into all Galileo platforms, including the new Galileo Protect and Evaluate. These tools leverage Luna's capabilities to intercept harmful inputs, improve system security, and increase operational efficiency. From Fortune 50 CPG brands to Fortune 10 US banks, teams are already using Luna to process millions of GenAI application queries per month, protect against malicious prompt injections, and reduce costs associated with GenAI operations.
read more: Amazon unveils new 'private investigator' AI model
[To share your insights with us as part of editorial or sponsored content, please write to psen@itechseries.com]
