Within days of launching, OpenAI’s chatbot, “ChatGPT,” went viral andquickly became a media fixture. Gaining notoriety and influence, ChatGPThas become the most successful example of the power of large languagemodels (LLMs)—the platform acquired 100 million active users in 2months, a feat Instagram took 2.5 years to achieve. Trained on a vast setof knowledge,1 and built on a Generative Pre-trained Transformer (GPT),OpenAI’s chatbot revolutionized the usability of GPT models. By dramaticallyimproving the underlying model’s generalizability and making it accessiblein an easy-to-use interface, OpenAI has sparked a seemingly limitless levelof enthusiasm and excitement about artificial intelligence (AI). With thepotential to address a vast array of downstream use cases, ChatGPT servesas an exemplar for future GPT implementations where the possibilities seemendless but may face unexpected challenges. As with any new technology,organizations will need the right skills, experience, and perspective to bestrealize the value of this kind of innovation. To inform such perspectives, weoffer the following informational white paper to explain what this technologyis and how organizations can use it.
However, to use ChatGPT effectively, users must first understand what it can and cannot do, as described in Figure 3. produce outputs that users can directly integrate into codebases, databases, and various components of a system, making the LLMs significantly more useful. There are four principles for prompts that every user should consider: Show and tellbe very clear in your instructions. Break work into smaller, discrete chunks. 3 Provide high quality datagive good examples and proofread your prompt instructions. Prompt it to check and improve its
id: b9dee92380c651fe4e8fee2dc7c3b0cb - page: 3
In addition to these principles, there are other useful guidelines to bear in mind. To mitigate the possibility that ChatGPT invents an incorrect response for a question about facts, OpenAI suggests that users: Provide a ground truth in a query. If you provide ChatGPT with a body of text to answer questions about (like a Wikipedia entry), it will be less likely to make up a response. Show ChatGPT how to say I dont know. If the model understands that in cases where its less certain about a response it can say I dont know, it will be less inclined to make up answers. High level principles and guidelines are helpful, but users who more deeply appreciate the kinds of prompt characteristics to consider will inevitably generate better results from ChatGPT. Additional conventions include, but are not limited to, the following: Explicitly state your desired focus, format, style, intended audience, and text length. Create a list of topics or points you want covered.
id: 25e8f9048de8797d42b8799c0a521a91 - page: 4
State the perspective from which you want the text written, if applicable. Specify requirements, like Add as many quantifiable references as possible. Break up the ask into smaller pieces; in this way, you can be more specific for long-form content. Adopt a coders mindset when creating promptsyou are programming a machine, not conversing with a human. 4 HOW WILL BUSINESSES AND ORGANIZATIONS USE CHATGPT? ChatGPTs capabilities transcend industries. Technology companies will not be the only organizations using ChatGPT, as it can provide value to a wide range of fields such as software development, healthcare, government and policy, military operations, and law.
id: 931779957ad843b5a7c963029af55f38 - page: 4
Potential Use Cases in the Government While the government is not typically an early adopter of new technology, it has several use cases similar to industry for proactively assessing and evaluating the benefits of employing a ChatGPT-like technology. Most use cases that involve highly repetitive tasks or processing large quantities of information can benefit from ChatGPT, and the government, like any organization, can realize substantial value in automating or accelerating all or parts of these tasks. However, ChatGPT is best used as a tool to augment the completion of human tasks rather than replace human workers. Some potential use cases in the government include: Figure 4: Examples of Potential Government Use Cases
id: b3c5c04b21700f693b21a2e62bb7886c - page: 4