AHvos, an innovator in enterprise-level, highly scalable AI solutions tailorable for data-heavy businesses, proudly presents a basic benchmark for a new AI approach to natural language understanding, a subset of natural language processing (NLP). 

This new AI procedure uses AHvos Corporation’s proprietary CRI (“Contextually Responsive Intelligence”) AI technology. CRI does not use any large language model (LLM) or any other type of neural network model. Compared to LLMs or other neural network-based techniques, CRI engines are educated with about 75% less data. The benchmark results show that CRI engines match, and in some instances, outperform state-of-the-art neural network-based NLP technology.

The CRI benchmark used 4 criteria to measure against for comparison purposes: accuracy, speed, hardware requirements and data required to train and educate. The CRI benchmark is comprised of two phases: the Education Phase and the Text Classification Phase. Comparatively, neural network-based NLPs utilize three phases: (1) the English-based AI large language model (LLM) generation, (2) the fine-tuning of the AI model for text classification, and (3) the text classification. 

The benchmark starts with an AHvos AI CRI engine that has no knowledge or pre-training of any kind related to any text language and it ends with a binarized text classification of the English subset of The Multilingual Amazon Reviews Corpus (MARC - published by the Association for Computational Linguistics). 

AHvos’ CRI showed an accuracy of 91.93%, as opposed to reported neural network-based AI accuracy levels of 52% to 73% for the English subset of MARC. The total AHvos CRI benchmark elapsed time was 1.8 seconds. In contrast, neural network-based AI benchmark published results required 10 hours just for fine-tuning and evaluation phases. While the time to generate popular LLM models is not publicly known, different independent studies estimate that they take between 6 months to 2.5 years. 

Lastly, the benchmark for CRI was conducted using a regular consumer laptop, while the published neural network-based AI benchmarks were run on an AWS p3.8xlarge instance or equivalent AI-specialized hardware. The benchmark hardware requirements demonstrated that alternative AI approaches that do not use neural networks are much more energy efficient thus enabling sustainable, green energy friendly AI.

“The results captured in this benchmark exercise demonstrate that our AHvos CRI engines are well suited for NLP tasks such as text classification,” says  Kevin Killens, CEO of AHvos. “The results of the benchmark prove that AHvos’ CRI provides a clear linguistic rationale for each of its text classification decisions resulting in a significant improvement in the understanding of AI classification responses versus conventional neural network-based technology.”

Further research has been completed by the Research and Development Department at AHvos Corporation to demonstrate that CRI engines are equally valid when using non-English datasets included in the Amazon Review Corpus as well as other types of text-based datasets producing both similar and improved accuracy and performance results.

AHvos launched its cutting-edge platform and its three initial market offerings in October 2024. Their private non-neural network-based AI engines, hosted by strategic partner Trinsic Technologies, avoid built-in biases and potential copyright infringement issues commonly found in AI models trained with open web content. AHvos’ technology replaces traditional point-in-time machine learning (ML) training with continual, near-real-time machine educating (ME).  

For more information about AHvos services, please visit www.ahvos.com.