Prompt chaining for LLM integrations in spreadsheet environments
Technology is disclosed herein for an application service which interfaces with an LLM service. In an implementation, the application service receives a natural language input from a user associated with a spreadsheet. The application service generates multiple prompts based on the natural language input and a portion of the spreadsheet and determines an order in which to input the prompts to an LLM service. The application service determines an order in which to input the prompts to an LLM service, then inputs the prompts to the LLM service according to the determined order.
1 . A computing apparatus comprising:
one or more computer-readable storage media;
one or more processors operatively coupled with the one or more computer-readable storage media; and
an application comprising program instructions stored on the one or more computer-readable storage media that, when executed by the one or more processors, direct the computing apparatus to at least:
receive, via a user interface of the application, a natural language input associated with a user with of a spreadsheet;
generate and submit multiple prompts to a large language model (LLM) service, wherein an initial prompt of the multiple prompts comprises the natural language input and a portion of the spreadsheet, wherein the initial prompt tasks the LLM service with generating a suggested formula responsive to the natural language input and returning a first output comprising the suggested formula in a parse-able format, and wherein a second prompt of the multiple prompts tasks the LLM service with generating a description of the suggested formula and returning a second output comprising the description of the suggested formula in the parse-able format;
store the first output and the second output received from the LLM service in a data structure associated with the spreadsheet;
parse the first output according to the parse-able format to extract the suggested formula for display in the user interface;
parse the second output according to the parse-able format to extract the description of the suggested formula and display, via the user interface, a graphical input device to display the description of the suggested formula;
receive, via the user interface, user input comprising a selection of the graphical input device; and
enable, via the user interface, display of the description of the suggested formula.
2 . The computing apparatus of claim 1 , wherein the parse-able format comprises one or more of: semantic tags and JavaScript Object Notation.
3 . The computing apparatus of claim 2 , wherein the second prompt is sent to the LLM service after the initial prompt.
4 . The computing apparatus of claim 3 , wherein a third prompt of the multiple prompts is sent after the second prompt, and wherein the third prompt comprises a request for the LLM service to output an explanation of the formula.
5 . The computing apparatus of claim 1 , wherein the program instructions further direct the computing apparatus to display replies by the LLM service in a progressive series of conversational dialogs in response to input by the user to proceed through the replies on a per-reply basis.
6 . The computing apparatus of claim 5 , wherein the program instructions direct the computing apparatus to input the multiple prompts to the LLM service prior to receiving any of the replies from the LLM service.
7 . The computing apparatus of claim 1 , wherein to input the multiple prompts to the LLM service, the program instructions direct the computing apparatus to input each prompt of the multiple prompts in-series to the LLM service.
8 . The computing apparatus of claim 7 , wherein to input the multiple prompts to the LLM service, the program instructions direct the computing apparatus to wait for a reply to a current one of the multiple prompts before inputting a next one of the multiple prompts.
9 . A method comprising:
receiving, via a user interface of an application, a natural language input associated with a user of a spreadsheet;
generating and submitting multiple prompts to a large language model (LLM) service, wherein an initial prompt of the multiple prompts comprises the natural language input and a portion of the spreadsheet, wherein the initial prompt tasks the LLM service with generating a suggested formula responsive to the natural language input and returning a first output comprising the suggested formula in a parse-able format, and wherein a second prompt of the multiple prompts tasks the LLM service with generating a description of the suggested formula and returning a second output comprising the description of the suggested formula in the parse-able format;
storing the first output and the second output received from the LLM service in a data structure associated with the spreadsheet;
parsing the first output according to the parse-able format to extract the suggested formula for display in the user interface;
parsing the second output according to the parse-able format to extract the description of the suggested formula and displaying, via the user interface, a graphical input device to display the description of the suggested formula;
receiving, via the user interface, user input comprising a selection of the graphical input device; and
enabling, via the user interface, display of the description of the suggested formula.
10 . The method of claim 9 , wherein a first one of the prompts comprises a request for the LLM service to output a formula for the spreadsheet the parse-able format comprises one or more of: semantic tags and JavaScript Object Notation.
11 . The method of claim 10 wherein the second prompt is sent to the LLM service after the initial prompt.
12 . The method of claim 11 , wherein a third prompt of the multiple prompts is sent after the second prompt, and wherein the third prompt comprises a request for the LLM service to output an explanation of the formula.
13 . The method of claim 9 , further comprising displaying replies by the LLM service in a progressive series of conversational dialogs in response to input by the user to proceed through the replies on a per-reply basis.
14 . The method of claim 13 , wherein inputting the multiple prompts to the LLM service occurs prior to receiving any of the replies from the LLM service.
15 . One or more computer-readable storage media having program instructions stored thereon that, when executed by one or more processors operatively coupled with the one or more computer-readable storage media, direct a computing device to:
receive, via a user interface of an application, a natural language input associated with a user of a spreadsheet;
generate and submit multiple prompts to a large language model (LLM) service, wherein an initial prompt of the multiple prompts comprises the natural language input and a portion of the spreadsheet, wherein the initial prompt tasks the LLM service with generating a suggested formula responsive to the natural language input and returning a first output comprising the suggested formula in a parse-able format, and wherein a second prompt of the multiple prompts tasks the LLM service with generating a description of the suggested formula and returning a second output comprising the description of the suggested formula in a parse-able format;
store the first output and the second output received from the LLM service in a data structure associated with the spreadsheet;
parse the first output according to the parse-able format to extract the suggested formula and display, via the user interface, a graphical input device to display the description of the suggested formula;
receive, via the user interface, user input comprising a selection of the graphical input device; and
display, via the user interface, the description of the suggested formula.
16 . The one or more computer-readable storage media of claim 15 , wherein a first one of the prompts comprises a request for the LLM service to output a formula for the spreadsheet the parse-able format comprises one or more of: semantic tags and Javascript Object Notation.
17 . The one or more computer-readable storage media of claim 16 , wherein the second prompt is sent to the LLM service after the initial prompt.
18 . The one or more computer-readable storage media of claim 17 , wherein a third prompt of the multiple prompts is sent after the second prompt, and wherein the third prompt comprises a request for the LLM service to output an explanation of the formula.
19 . The one or more computer-readable storage media of claim 15 , wherein the program instructions further direct the computing device to display replies by the LLM service in a progressive series of conversational dialogs in response to input by the user to proceed through the replies on a per-reply basis.
20 . The one or more computer-readable storage media of claim 19 , wherein the program instructions direct the computing device to input the multiple prompts to the LLM service prior to receiving any of the replies from the LLM service.