Despite his age, Adarsh Shirawalmath, the creator of Kannada Llama and founder of Tensoic, shared on LinkedIn that he reached out to Greg Brockman from OpenAI in 2020 to request access to GPT-3, and he received it.
What’s intriguing is his aim to use GPT-3 to predict JEE exam questions, a detail he mentioned to Brockman.
Back then, only those deeply involved in AI/ML knew about GPT-3. Shirawalmath found the concept of predicting the next words intriguing, believing that ‘predicting anything’ seemed achievable,” Shirawalmath stated.
Although he had previously applied for API access, due to exam preparations, he contacted Brockman for early model access and surprisingly received it within days. He had mentioned this to AIM on an episode of Tech Talks.
Reflecting on the experience, Shirawalmath believes it may have been one of the first instances of few-shot prompting and RAG. By inputting questions from previous year papers and asking for predictions of the next set of questions, he found some success, but did not save the results.
However, the JEE question paper predictor repository remains available. Visit here to explore it.
AIM tested ChatGPT and Ola Krutrim on their ability to pass the UPSC Prelims exam as well.
