Recently, Odia Generative AI introduced an AI mentor named Acharya tailored for the CBSE Board.
A year ago, when ChatGPT was first introduced, Shantipriya Parida, the mind behind Odia Llama, felt a sense of disappointment as it failed to grasp any cultural nuances related to Odisha.
Fast forward to the present, Parida has developed Odia Llama (Llama2 – a fine-tuned LLM for the Odia language) and initiated an open-source initiative named Odia Generative AI. With over 35 million speakers, Odia boasts a rich literary heritage and distinctive identity.
“I attempted to inquire about local contexts from ChatGPT. For instance, I asked, ‘Can you provide the recipe for rasgulla?’ It was unable to furnish a response. It lacked the understanding of local nuances, resulting in inaccurate answers,” Parida recalled.
Currently employed as a senior AI scientist at Silo.ai in Finland, which recently unveiled its own LLM named Poro, Parida remarked, “If Europeans can develop an LLM that outperforms OpenAI’s ChatGPT in their native language, why can’t we achieve the same for our Indic languages? That’s what drives our relentless efforts, and it’s gratifying to receive acknowledgment for our endeavors.”
The Path of Odia Llama
To construct Odia Llama, Parida devised a three-phase strategy. “We’ll commence with a fine-tuned model, progress to a pre-trained model, and ultimately, proceed to application deployment. Once the fine-tuned and pre-trained models are ready, app development will be relatively straightforward,” he elucidated.
Prasanta Patnaik