AI21 Labs has recently introduced "Contextual Answers," a question-answering engine designed for large language models (LLMs). This new engine, when connected to an LLM, allows users to upload their own databases, enabling the model's output to be limit ed to specific information. The launch of chatbots like ChatGPT and other AI products has revolutionized the AI industry, but credibility concerns have hindered widespread adoption in various businesses.
Research indicates that employees spend nearly half of their working day searching for information, presenting a significant opportunity for chatbots capable of performing search functions. However, most chatbots are not suitable for businesses. Contextual Answers aims to bridge the gap between general-purpose chatbots and enterprise-grade Q&A services by empowering users to manage their data and document repositories.
According to AI21's blog post, Contextual Answers addresses key barriers to adoption by allowing users to utilize AI answers without the need for retraining the model. Many businesses struggle to adopt AI due to factors like cost, complexity, and lack of expertise in integrating Models with their organizational data, leading to incorrect responses, "hallucinations," or irrelevant context.
One of the challenges in developing useful LLMs is teaching them to express a lack of confidence. When faced with queries for which they lack sufficient information, chatbots often fabricate answers without factual basis instead of responding with a low-confidence answer like "I don't know." AI21 claims that Contextual Answers solves this problem by outputting information only when relevant to the user's provided document or not outputting any information at all, thereby eliminating hallucination issues.
In industries like finance and law, where accuracy is critical, the introduction of generative pre-trained transformer (GPT) systems has yielded mixed results. Experts remain cautious about using GPT systems in the financial community due to their potential to hallucinate or produce misleading information, even when connected to the internet and linked to sources. Similarly, in the legal space, there have been instances of lawyers facing fines and sanctions for relying on ChatGPT-generated outputs in court cases.
With Contextual Answers, AI21's approach of preloading relevant data into AI systems and intervening before non-factual information is produced seems promising in mitigating the hallucination problem. This development could lead to broader adoption, especially in the fintech sector, where Traditional financial institutions have been hesitant to embrace GPT technology, while the cryptocurrency and blockchain communities have experienced varying success in leveraging chatbots.
