Krutrim unveils its Language Model Models (LLMs), including the Krutrim base and Krutrim Pro versions. The Pro variant is highlighted as a multimodal foundational model,
Krutrim, led by founder Bhavish Aggarwal, is set to launch its first product, an innovative AI chatbot application. The startup aims to transform digital interactions with its advanced technology. Aggarwal announced the upcoming launch on social media platform X (formerly Twitter), providing insights into the capabilities of the AI offering through shared screenshots.
The startup asserts that all forthcoming AI products will demonstrate an impressive comprehension of India’s diverse linguistic landscape, covering 22 scheduled languages. Moreover, these innovative offerings will feature content-generation capabilities spanning eight languages.
Aggarwal’s announcement on February 4 hinted at the final stages of testing for the @Krutrim app, slated for release in the upcoming week. While expressing satisfaction with the current state of the app, Aggarwal pledged ongoing enhancements to the AI models even after launch.
In a series of posts, Aggarwal showcased the chatbot’s versatile capabilities, demonstrating its adeptness in handling basic inquiries and offering insightful suggestions. However, one specific interaction garnered substantial scrutiny from netizens.
Upon querying the Krutrim AI chatbot about India’s status as a country before independence, the app’s response ignited controversy. It claimed that India existed as a sovereign entity before the British Raj, sparking debate among users. Aggarwal compared this response with that of OpenAI’s ChatGPT, which presented a different perspective on India’s historical context.
Critics accused the Krutrim AI chatbot of displaying signs of “AI hallucination,” where large language models generate inaccurate or fictional information. Many users favored ChatGPT’s response, citing its alignment with historical accuracy. Concerns were also raised about a user interface flaw, particularly the misalignment of the ‘Send’ icon with the text box.
Krutrim has revealed information about its Language Model Models (LLMs), which include the Krutrim base and Krutrim Pro variants. The latter is highlighted as a multimodal foundational model. These models have been trained extensively on a corpus containing over two trillion Indic language tokens. Although Krutrim has hinted at voice-enabled functionality for its generative AI apps, it’s unclear if this feature will be included at launch.
The company has shared performance benchmarks of its AI model, comparing it with Meta Llama 2 7B, a coding assistant model trained on over 500 billion tokens. While Krutrim outperformed Meta Llama 2 7B in benchmarks like Arc, Hellaswag, and Copa, it lagged behind in benchmarks such as MMLU, Bigbench, and Lambada_open.
