OpenAi's product chatGPT swept the world in late 2022 and was the first time interacting with large models like GPT-3 for many. The simple easy of asking for something like you would send a text message was an immediate hit and something almost all other software is not trying to emulate in some way.
How can I use chatGPT but with my own files?
Everyone quickly started to ask, how can I use something like chatGPT but with my own files, data, or information instead of using general knowledge of the whole internet.
Finding the right piece of information is hard. Even with perfect file organization, almost anyone who works with even moderate sized information spends time searching for something, especially if they don't know exactly what file is most relevant.
Semantic search
One of the breakthroughs that lead to the availability of models trained on billions of lines of internet writing has popularized and made possible a type of information retrieval called semantic search.
While the mechanics are complicated, this essentially means that the search is able to match similar strings of text like Apple and Fruit, or Car and Tire. This is done by converting each word, regardless of language, into mathematical representation that can be plotted on a graph where you can detect the distance between various words and meanings.
Find it in the fourth cabinet on the left
This search technique can be used to power a chatbot like chatGPT to locate and return similar text strings to what you search.
This can then be combined with other AI models to produce answers to your questions relevant to specific blocks of text or files.
Effectively, you are giving a model like chatGPT the means to find and surface source material about the question that you asked.
Fewer tables and more insights
While document search is powerful, more generally, this technology unlocks new and powerful ways of analyzing information.
Missing the forest for the trees
We have used a similar framework to power PromptLoop models to return the correct information, wether categorizing text or generating it.
Data analysis, especially the analysis carried out in spreadsheet models, has evolved based on exact matches to IDs, numbers or tables. This has led to the need to have exact formulas and often times make messy text much harder to use than it should be.
Smart matching
One of the powerful ways PromptLoop is used in information analysis like processing form responses or customer lists is by assigning each row to a specific category.
PromptLoop Label uses the same information retrieval techniques that can be combined with chatGPT to identify the correct category match to label each row.
This can then be used in a PivotTable or any other summary tool used in the workbook. Instead of getting #N/A errors or false matches you can drop each item into its appropriate bucket.
Quantifying text inputs
Another way to use this information retrieval technique is through converting text data into easy to use numbers.
As an example, a customer was able to take short form survey responses, and convert them to a 1-10 net promoter score that could then be quickly graphed and analyzed as a trend.
This works similarly to above function but allows you to match to example number scales instead of descriptive categories which can be equally powerful.
Bring chatGPT to your data
Right now, everyone should think of AI models as tools to be used where they can be most helpful. For most of us, our company and work data is what we spend the most time thinking about.
Using models to process and surface this information can be as effective as it is easy to implement with tools like PromptLoop and chatGPT. Get started with a free trial today or explore out templates