Background Over the last year, Large language models (LLMs) such as ChatGPT, Gemini and Llama have shown remarkable intelligence capabilities and are increasingly being used in a variety of applications. However, off-the-shelf LLMs do not suffice for all use cases. In case of open source LLMs such as Llama, one option is to fine-tune the LLM for the specific use case or task; however, this can often be prohibitively expensive. Further, Responsible AI (RAI) concerns such as safety and toxicity, and their tradeoffs with model accuracy can be different for different applications, making a one-size-fits-all approach less desirable.
LLM-based systems and Responsible AI
LLM-based systems and Responsible AI
LLM-based systems and Responsible AI
Background Over the last year, Large language models (LLMs) such as ChatGPT, Gemini and Llama have shown remarkable intelligence capabilities and are increasingly being used in a variety of applications. However, off-the-shelf LLMs do not suffice for all use cases. In case of open source LLMs such as Llama, one option is to fine-tune the LLM for the specific use case or task; however, this can often be prohibitively expensive. Further, Responsible AI (RAI) concerns such as safety and toxicity, and their tradeoffs with model accuracy can be different for different applications, making a one-size-fits-all approach less desirable.