Contextual AI’s new AI mannequin crushes GPT-4o in accuracy — right here’s why it issues

Contextual AI unveiled its grounded language mannequin (GLM) at the moment, claiming it delivers the very best factual accuracy within the {industry} by outperforming main AI techniques from Google, Anthropic and OpenAI on a key benchmark for truthfulness.

The startup, based by the pioneers of retrieval-augmented technology (RAG) know-how, reported that its GLM achieved an 88% factuality rating on the FACTS benchmark, in comparison with 84.6% for Google’s Gemini 2.0 Flash, 79.4% for Anthropic’s Claude 3.5 Sonnet and 78.8% for OpenAI’s GPT-4o.

Whereas massive language fashions have remodeled enterprise software program, factual inaccuracies — typically known as hallucinations — stay a crucial problem for enterprise adoption. Contextual AI goals to resolve this by making a mannequin particularly optimized for enterprise RAG purposes the place accuracy is paramount.

“We knew that part of the solution would be a technique called RAG — retrieval-augmented generation,” mentioned Douwe Kiela, CEO and cofounder of Contextual AI, in an unique interview with VentureBeat. “And we knew that because RAG is originally my idea. What this company is about is really about doing RAG the right way, to kind of the next level of doing RAG.”

The corporate’s focus differs considerably from general-purpose fashions like ChatGPT or Claude, that are designed to deal with all the pieces from artistic writing to technical documentation. Contextual AI as an alternative targets high-stakes enterprise environments the place factual precision outweighs artistic flexibility.

“If you have a RAG problem and you’re in an enterprise setting in a highly regulated industry, you have no tolerance whatsoever for hallucination,” defined Kiela. “The same general-purpose language model that is useful for the marketing department is not what you want in an enterprise setting where you are much more sensitive to mistakes.”

How Contextual AI makes ‘groundedness’ the brand new gold normal for enterprise language fashions

The idea of “groundedness” — guaranteeing AI responses stick strictly to info explicitly supplied within the context — has emerged as a crucial requirement for enterprise AI techniques. In regulated industries like finance, healthcare and telecommunications, corporations want AI that both delivers correct info or explicitly acknowledges when it doesn’t know one thing.

Kiela supplied an instance of how this strict groundedness works: “If you give a recipe or a formula to a standard language model, and somewhere in it, you say, ‘but this is only true for most cases,’ most language models are still just going to give you the recipe assuming it’s true. But our language model says, ‘Actually, it only says that this is true for most cases.’ It’s capturing this additional bit of nuance.”

The power to say “I don’t know” is a vital one for enterprise settings. “Which is really a very powerful feature, if you think about it in an enterprise setting,” Kiela added.

Contextual AI’s RAG 2.0: A extra built-in solution to course of firm info

Contextual AI’s platform is constructed on what it calls “RAG 2.0,” an strategy that strikes past merely connecting off-the-shelf elements.

“A typical RAG system uses a frozen off-the-shelf model for embeddings, a vector database for retrieval, and a black-box language model for generation, stitched together through prompting or an orchestration framework,” based on an organization assertion. “This leads to a ‘Frankenstein’s monster’ of generative AI: the individual components technically work, but the whole is far from optimal.”

As a substitute, Contextual AI collectively optimizes all elements of the system. “We have this mixture-of-retrievers component, which is really a way to do intelligent retrieval,” Kiela defined. “It looks at the question, and then it thinks, essentially, like most of the latest generation of models, it thinks, [and] first it plans a strategy for doing a retrieval.”

This complete system works in coordination with what Kiela calls “the best re-ranker in the world,” which helps prioritize essentially the most related info earlier than sending it to the grounded language mannequin.

Past plain textual content: Contextual AI now reads charts and connects to databases

Whereas the newly introduced GLM focuses on textual content technology, Contextual AI’s platform has not too long ago added help for multimodal content material together with charts, diagrams and structured knowledge from well-liked platforms like BigQuery, Snowflake, Redshift and Postgres.

“The most challenging problems in enterprises are at the intersection of unstructured and structured data,” Kiela famous. “What I’m mostly excited about is really this intersection of structured and unstructured data. Most of the really exciting problems in large enterprises are smack bang at the intersection of structured and unstructured, where you have some database records, some transactions, maybe some policy documents, maybe a bunch of other things.”

The platform already helps quite a lot of advanced visualizations, together with circuit diagrams within the semiconductor {industry}, based on Kiela.

Contextual AI’s future plans: Creating extra dependable instruments for on a regular basis enterprise

Contextual AI plans to launch its specialised re-ranker element shortly after the GLM launch, adopted by expanded document-understanding capabilities. The corporate additionally has experimental options for extra agentic capabilities in growth.

Based in 2023 by Kiela and Amanpreet Singh, who beforehand labored at Meta’s Elementary AI Analysis (FAIR) staff and Hugging Face, Contextual AI has secured clients together with HSBC, Qualcomm and the Economist. The corporate positions itself as serving to enterprises lastly notice concrete returns on their AI investments.

“This is really an opportunity for companies who are maybe under pressure to start delivering ROI from AI to start looking at more specialized solutions that actually solve their problems,” Kiela mentioned. “And part of that really is having a grounded language model that is maybe a bit more boring than a standard language model, but it’s really good at making sure that it’s grounded in the context and that you can really trust it to do its job.”