Contextual AI’s new AI model crushes GPT-4o in accuracy — here’s why it matters

You have context It was presented The grounded language model (GLM) claims that today, the highest actual accuracy of the industry, which is superior to leading AI systems Google, Anthropical and Open over the main criterion for the truth.

Start-established by pioneers Returned generation (Dwarf) technology said that 88% of GLM would actually score a point in fact Facts BenchmarkCompared to 84.6% for Google Twins 2.0 Flash79.4% for anthropic Claude 3.5 Sonnet and 78.8% for Openai GPT-4O.

Extensive language models are a transformprice program, actual inaccuracies – often called hallucinations – remains a critical challenge for business reception. Contextual AI aims to solve this by creating a specially optimized model for enterprise dwarf applications where accuracy is paramount.

“We knew that part of the solution will be a tech called a ragged generation,” he said. “And we knew that the cloth is my true opinion. What is the company, indeed, to the next level of the fluttering level.”

The company’s focus differs significantly from generalial models Chatgpt or ClaudThe creative writing is designed to manage everything to technical documents. Instead of a contextual AI, the actual accuracy target the high-step enterprise environment where the creative flexibility is superior.

“If you have a diaper problem and are within a highly adjustable industry, there is no endurance for the halla,” Kiela said. “The same general language model that is useful for the marketing department is not more sensitive to errors in a venture.”

Contextual AI is a new gold standard ‘based’ for enterprise language models

The concept “reason“Serious gluing of the information provided in the context, the entity has a critical requirement for the AI ​​systems.

Kieta suggested a sample of how this strict reasonality works. However, our language model is said: “In fact, it is only in most cases that this is true.” This is the seizure of an additional nuance. “

The ability to say “I don’t know” is important for enterprise parameters. “If you really are a very strong feature, if you think about this when a venture setting,” said Kie.

Contextual Ai’s RAG 2.0: A more integrated way to process company data

Contextual Ai’s platform is built on what they call “RAG 2.0“Just an approach that goes beyond combining shelf components.

“A typical dwarf system uses a vector database for a vector database and a company that is a vector database for salvation.

Instead, the AI ​​contextual AI jointly optimizes all components of the system. “We have a component of this mixture, which is a way of searching for a smart,” said Kiela said. “It looks at this question, then thinks like most of the latest generation models, it thinks, [and] Originally planning a strategy for a search. “

This system works in the coordination of this system, called the “best record”, which helps to prioritize the most suitable information before sending Kieta to a substantiated language model.

Out of the plain text: Contextual AI now reads graphs and connects to the database

The newly announced GLM is aimed at the text generation, the contextual AI platform added support for multimodal content, including recently supported content and built from popular platforms Large, Avalanche, Redshift and Postgres.

“The most difficult problems in the enterprises are in the intersection of unstructured data,” said KieLa. “I’m mostly excited about what I am really structured and unstructured. In large enterprises, most of the exciting problems, some transactions, maybe some policy documents, perhaps a bunch of possibly, maybe a bunch, maybe a bunch of documents.

According to the platform, Kieela supports various complex visual images, including circuit diagrams in the semiconductor industry.

Contextual AI’s Future Plans: Create more reliable tools for daily work

Contextual AI plans to release a reckonent in a short time after the start of GLM, and the capacity of the expanded documentation. The company also has experimental features for more agency opportunities in development.

Kieta and in 2023 Amanpreet singhPreviously, the Foundation of Methadamal AI research (fair) has provided its customers, including the facial, contextual EU HSBC, Qualcomm and Economist. The company manifests itself as helps to enterprises, finally realizes specific revenues in EU investments.

“This is an opportunity for companies that are under pressure to start delivering Roi from AI to start looking at more specialized solutions,” said Kie. “And part of this has a reasoned language model that is a slightly boring language model, but it is really good to make sure you can trust it in the context and to see his work.”