With Laya and Sol, Articul8 puts language scholars at heart of heritage AI

Articul8, an artificial intelligence company, has introduced a heritage AI model called Laya, which prioritizes linguistic purity and heritage. The Laya AI model is based on the ancient Sanskrit code of 3,996 interlocking grammatical rules, which guides its responses and refusal to agree with incorrect interpretations. The model was unveiled by Union Finance Minister Nirmala Sitharaman on September 15 at the 118-year-old Madras Sanskrit College, where it is currently undergoing extended testing before broader public availability.

Laya, a 32-billion-parameter model, is designed to stand its ground and refuse to answer when faced with incorrect interpretations, unlike most large language models that typically accept and agree with flawed generations. This unique feature was demonstrated when Sanskrit scholars presented Laya with an incorrect interpretation of a verse, and it refused to agree based on its core knowledge base.

Articul8's CEO, Arun Subramaniyan, highlighted the ambition behind this heritage AI initiative, which now extends to Tamil with its "Sol" model. The Sanskrit Heritage Model and Platform uphold linguistic purity by tracing every output directly to its source text, ensuring that outputs are rooted in the original text. By enforcing grammatical laws codified in Panini’s Ashtadhyayi over two millennia ago, Laya and Sol reject flawed generations and promote concepts unique to Indian culture and tradition.

The five-part architecture of Laya includes a custom tokeniser for Indic languages, a model that breaks down queries into individual words, a context embedding model to handle different meanings of the same word, a rule verification engine to ensure the right set of rules are being used, and the Laya generation model itself. Subramaniyan emphasized that enterprises and users can use other models alongside Laya, but the advantage lies in Laya's efficiency, accuracy, and adherence to grammatical laws.

Plans are underway to extend the language heritage model to other Indian languages, such as Gujarati, Kannada, and Telugu, leveraging the successful framework used for Tamil. By exploring older texts for insights into various fields like astronomy, metallurgy, and sustainability, Articul8 aims to uncover potential commercial discoveries that could benefit society at large.

Articul8 is in the process of finalizing a partnership with a central government institution, which would validate and jointly release Sol. Subramaniyan emphasized the importance of respecting the custodians of the language and sees these efforts as a necessary defense measure for nations in the AI era.