The internet has been aflame with the fun and fear of the uncannily human ChatGPT demo from OpenAI. This model was optimized for the format (chat) and to be as human as possible (doc). However, this does not mean it’s been optimized for utility. Before the launch of ChatGPT, OpenAI’s core models have been powering deeper task-oriented applications. Particularly the DaVinci GPT-3 Model is the highest quality version that enables a user to unlock the true power of the GPT-Model (available via the playground).
In Davinci’s own words
How should i use Chat-GPT vs text-Davinci? (from text-davinci-003 model):
Chat-GPT is a natural language processing (NLP) tool that can be used to generate text-based conversations. It is designed to be used in chatbot applications, where it can generate responses to user queries. Text-Davinci is an AI-powered text editor that can be used to create and edit text documents. It is designed to be used for writing and editing text documents, and can be used to generate text-based conversations. Both tools can be used to generate text-based conversations, but Chat-GPT is better suited for chatbot applications, while Text-Davinci is better suited for writing and editing text documents.
Before diving into the use cases let’s examine some of the system limitations/strengths to guide use to the best applications.
Limitations
1 — Doesn’t synthesize information well. The models are fairly unreliable when summarizing or providing options on data sets. It’s best to aggregate facts with sources cited so you can confirm.
2 — Using data only up to 2021. This makes it great for general information about the world, but for finding contemporary data (news/stocks) it’s lacking.
Strengths
1 — Formatting — models can format in complex ways that would be time intensive for humans.
2 — Search in parallel. Often searching is a process requiring multiple searches of high levels of detail. the GPT models can do this all in parallel for you.
3 — Writing ins prose. Best known application — when you want something human readable the GPT models lower the cost of production here to near 0.
Most Useful Applications (today)
Shopping
The AI Models are great at quickly searching the web, parsing out data and returning it in an easy to use format. Even for niche applications (ie surf schools in Haleiwa) it can quickly find the information that would have taken a human several minutes of searching/copying/pasting to collect:
Market Research
Just as with shopping DaVinci excels at taking complex/disparate data and filtering it down into a chart. Note — it doesn’t always understand the context, so be sure to check it’s work here.
Ideation
Often we struggle with blank-page problems where it’s hard to get started. Good news is DaVinci can get creative processes kicked off. Whether it’s a name or a topic sentence the models can help get the process kicked off. This can also be applied to creating ad title/copy combinations to test conversion rates with.
Writing Prose
In many situations the information to be conveyed is clear, but the social context requires prose to be written around that. DaVinci can quickly produce prose to save you the work of doing it yourself.
Summarization
In a more limited way DaVinci can summarize text. This has been one of the least reliable areas of the tool, but one with the most promise.
Application Themes
As you consider applications beyond the examples listed focus on tasks that are verifiable but time intensive. Given the AI doesn’t really have the capability for judgement or context it needs very specific parameters that you can verify (and potential edit/filter for it) in order to be useful.
