According to the business, Krutrim AI’s chatbot is capable of comprehending all 22 scheduled languages spoken in India.
The first product from Krutrim Si Designs, also known as Krutrim, is a generative artificial intelligence (AI) chatbot app that will soon be released. The startup’s creator, Bhavish Aggarwal, used X (previously known as Twitter) to announce the first AI product’s launch schedule and to release screenshots showcasing some of its features. The business claims that all of its AI devices would be able to comprehend India’s 22 scheduled languages. Furthermore, the items would be able to produce content in eight different languages.
On February 4, Aggarwal shared a post on his social media account saying, “Testing the @Krutrim app before release! release the following week. Although the models are good, we will continue to refine them after launch. He later demonstrated the chatbot’s creative skills in a number of posts.
A few articles demonstrated that the chatbot can respond to simple questions and even make recommendations, but one in particular attracted a lot of backlash from online users.
When the user asked the Krutrim AI chatbot if India existed prior to independence, the app replied, “Yes, India did exist prior to the British Raj.” Its political structure, culture, language, religion, and customs were all unique. Aggarwal captioned the image, saying, “This is why we need to build India’s own AI.” He also shared the ChatGPT response from OpenAI, which read, “Before the establishment of the British Raj, the region that is now known as India was a collection of various kingdoms, empires, and principalities.
In response, a number of users claimed that ChatGPT’s response was more accurate and that the post was an example of AI hallucination—a phenomena in which large language models (LLMs) produce erroneous or nonexistent information. Additionally, several users brought attention to a bug in which the text box’s empty spaces were not aligned with the “Send” symbol.
The AI startup claims to have developed the Krutrim basic and Krutrim Pro LLMs, the latter of which is referred to as a multimodal fundamental model. More than two trillion tokens in the Indian language were used to train the models. Additionally, the business has stated that speech support would be available for its generative AI apps, though it’s unclear if this functionality will be included at launch. The business has also shared the benchmark scores of their AI model on its website, contrasting it with the Meta Llama 2 7B model, a coding assistance that has been trained on over 500 billion tokens. Krutrim is shown as winning on the Arc, Hellaswag, Copa, Jeopardy, Piqa, and other benchmarks, according to the chart.
Source Link
