I find that a very useful and helpful framework to think about what is a good candidate for us to apply language models like chat GP data is when we are generating content. Whether it’s writing a blog post, marketing copy, e-commerce product description, or a sales email, we can ask ourselves if it’s cheaper to have a draft and correct it or create it from scratch. Another important consideration is whether we can live with the consequences if a mistake goes unnoticed and is published. If an application satisfies these considerations, it’s worth experimenting with language models.
One of the most mature applications of language models is in software development. Programmers worldwide have adopted the use of these tools to write software. Since language models can generate code, it’s much cheaper to have the technology write the code for you and then verify and test it. This has resulted in a significant increase in productivity for software development.
Another application of language models is in content generation. Companies are experimenting with using language models to write blog posts, marketing and sales content, and product descriptions. It’s cheaper to fix a language model’s draft than to write the entire content from scratch. However, it’s important to have a human in the loop to verify and ensure the accuracy and relevance of the generated content.
Language models are also being used in question answering applications. In an enterprise setting, where answers are buried in numerous documents, language models can provide a conversational search experience. Users can ask questions and receive answers, and even ask follow-up questions. However, it’s crucial to verify the accuracy of the answers as language models can occasionally provide incorrect information. Having a human in the loop and asking for the source document can help ensure the accuracy of the answers.
While language models have shown great potential in these applications, it’s important to be cautious. The risk of language models providing incorrect information to customers is a real concern. Therefore, rollouts of chatbots for external customers are more cautious and slow. However, for internal question answering, many companies have already implemented chatbots to improve productivity and access information within their organizations.
