fixed gptpopup to do something more reasonable when asked to tag documents.

2 files changed, 43 insertions, 31 deletions

diff --git a/src/client/apis/gpt/GPT.ts b/src/client/apis/gpt/GPT.ts
index 7878e9bfe..af9cdd2d1 100644
--- a/src/client/apis/gpt/GPT.ts
+++ b/src/client/apis/gpt/GPT.ts
@@ -11,10 +11,12 @@ export enum GPTDocCommand {
 
 export const DescriptionSeperator = '======';
 export const DocSeperator = '------';
+export const DataSeperator = '>>>>>>';
 export enum TextClassifications {
-    Title = 'word', //a few words 
+    Title = 'word', //a few words
     Caption = 'sentence', //few sentences
-    LengthyDescription = 'paragraphs' }
+    LengthyDescription = 'paragraphs',
+}
 
 enum GPTCallType {
     SUMMARY = 'summary',
@@ -38,11 +40,12 @@ enum GPTCallType {
     MAKERUBRIC = 'make_rubric', // create a definition rubric for a document to be used when quizzing the user
     COMMANDTYPE = 'command_type', // Determine the type of command being made (GPTQueryType - eg., AssignTags, Sort, Filter, DocInfo, GenInfo) and possibly some parameters (eg, Tag type for Tags)
     SUBSETDOCS = 'subset_docs', // select a subset of documents based on their descriptions
+    TAGDOCS = 'tag_docs', // choose a tags for each Doc
     DOCINFO = 'doc_info', // provide information about a document
     SORTDOCS = 'sort_docs',
     CLASSIFYTEXTMINIMAL = 'classify_text_minimal', // classify text into one of the three categories: title, caption, lengthy description
     CLASSIFYTEXTFULL = 'classify_text_full', //tags pdf content
-    GENERATESCRAPBOOK = 'generate_scrapbook'
+    GENERATESCRAPBOOK = 'generate_scrapbook',
 }
 
 type GPTCallOpts = {
@@ -55,14 +58,12 @@ type GPTCallOpts = {
 const callTypeMap: { [type in GPTCallType]: GPTCallOpts } = {
     // newest model: gpt-4
     summary: { model: 'gpt-4-turbo', maxTokens: 256, temp: 0.5, prompt: 'Summarize the text given in simpler terms.' },
-    
-    
+
     sort_docs: {
         model: 'gpt-4o',
         maxTokens: 2048,
         temp: 0.25,
-        prompt: 
-            `The user is going to give you a list of descriptions. 
+        prompt: `The user is going to give you a list of descriptions. 
                 Each one is separated by '${DescriptionSeperator}' on either side. 
                 Descriptions will vary in length, so make sure to only separate when you see '${DescriptionSeperator}'. 
                 Sort them by the user's specifications. 
@@ -71,7 +72,6 @@ const callTypeMap: { [type in GPTCallType]: GPTCallOpts } = {
                 It is VERY important that you format it exactly as described, ensuring the proper number of '${DescriptionSeperator[0]}' and '${DocSeperator[0]}' (${DescriptionSeperator.length} of each) and NO commas`,
     },
 
-    
     edit: { model: 'gpt-4-turbo', maxTokens: 256, temp: 0.5, prompt: 'Reword the text.' },
     stack: {
         model: 'gpt-4o',
@@ -100,7 +100,7 @@ const callTypeMap: { [type in GPTCallType]: GPTCallOpts } = {
         temp: 0.25,
         prompt: `Based on the content of the the text, classify it into the  
                 most appropriate category: '${TextClassifications.Title}' if it is a few words, '${TextClassifications.Caption}' if it is a couple sentences or less, or '${TextClassifications.LengthyDescription}' if it is a lengthy description. Output exclusively the classification in your response.
-                `
+                `,
     },
     classify_text_full: {
         model: 'gpt-4o',
@@ -109,7 +109,7 @@ const callTypeMap: { [type in GPTCallType]: GPTCallOpts } = {
         prompt: `Based on the content of the the text, classify it into the  
                 most appropriate category: '${TextClassifications.Title}', '${TextClassifications.Caption}', or '${TextClassifications.LengthyDescription}'. 
                Then provide five more descriptive tags (single words) separated by spaces. 
-            Finally, include a more detailed summary phrase tag using underscores, for a total of seven tags.`
+            Finally, include a more detailed summary phrase tag using underscores, for a total of seven tags.`,
     },
     describe: { model: 'gpt-4-vision-preview', maxTokens: 2048, temp: 0, prompt: 'Describe these images in 3-5 words' },
     flashcard: {
@@ -174,7 +174,7 @@ const callTypeMap: { [type in GPTCallType]: GPTCallOpts } = {
         model: 'gpt-4o',
         maxTokens: 2048,
         temp: 0.5,
-         prompt: `Generate an aesthetically pleasing scrapbook layout preset based on these items.
+        prompt: `Generate an aesthetically pleasing scrapbook layout preset based on these items.
     Return your response as JSON in the format:
     [{
         "type": rich text or image or pdf or video or collection
@@ -191,22 +191,32 @@ const callTypeMap: { [type in GPTCallType]: GPTCallOpts } = {
                       tag: 
                       x: , y: , width: , height:
                     }
-                  ] `
-
-      },
+                  ] `,
+    },
 
     command_type: {
         model: 'gpt-4-turbo',
         maxTokens: 1024,
         temp: 0,
-        prompt: `I'm going to provide you with a question. 
-                 Based on the question, is the user asking you to 
-                     ${GPTDocCommand.AssignTags}. Assigns docs with tags(like star / heart etc)/labels.
+        prompt: `Is the provided command/question asking you to:
+                     ${GPTDocCommand.AssignTags}. Choose a descriptive tag/label.
                      ${GPTDocCommand.GetInfo}. Provide information about a specific doc.
                      ${GPTDocCommand.Filter}. Filter docs based on a question/information.
                      ${GPTDocCommand.Sort}. Put docs in a specific order. 
-                Answer with only the number for ${GPTDocCommand.GetInfo}-${GPTDocCommand.Sort}. 
-                For number one, provide the number (${GPTDocCommand.AssignTags}) and the appropriate tag`,
+                Answer with only the number, do not add any other punctuation or formatting of any kind`,
+    },
+    tag_docs: {
+        model: 'gpt-4-turbo',
+        maxTokens: 1024,
+        temp: 0,
+        prompt: `I'm going to give you a list of descriptions.
+                Each one is separated by '${DescriptionSeperator}' on either side. 
+                Descriptions will vary in length, so make sure to only separate when you see '${DescriptionSeperator}'. 
+                Based on the question/command the user asks, provide a tag label of the given descriptions that best matches the user's specifications.
+                Format your response precisely as a single string that prints each description followed by '${DataSeperator}' followed by the label followed by '${DescriptionSeperator}'.
+                Do not use any additional formatting marks or punctuation'.
+                Immediately afterward, surrounded by '${DocSeperator}' on BOTH SIDES, provide some insight into your reasoning for the way you sorted (and mention nothing about the formatting details given in this description). 
+                `,
     },
     subset_docs: {
         model: 'gpt-4-turbo',
@@ -284,7 +294,7 @@ const gptImageCall = async (prompt: string, n?: number) => {
             n: n ?? 1,
             size: '1024x1024',
         });
-        return response.data.map((data: Image) => data.url);
+        return (response.data ?? []).map((data: Image) => data.url);
         // return response.data.data[0].url;
     } catch (err) {
         console.error(err);
diff --git a/src/client/views/pdf/GPTPopup/GPTPopup.tsx b/src/client/views/pdf/GPTPopup/GPTPopup.tsx
index 660284397..ec269cc6c 100644
--- a/src/client/views/pdf/GPTPopup/GPTPopup.tsx
+++ b/src/client/views/pdf/GPTPopup/GPTPopup.tsx
@@ -11,7 +11,7 @@ import { ClientUtils } from '../../../../ClientUtils';
 import { Doc } from '../../../../fields/Doc';
 import { NumCast, StrCast } from '../../../../fields/Types';
 import { Networking } from '../../../Network';
-import { DescriptionSeperator, DocSeperator, GPTCallType, GPTDocCommand, gptAPICall, gptImageCall } from '../../../apis/gpt/GPT';
+import { DataSeperator, DescriptionSeperator, DocSeperator, GPTCallType, GPTDocCommand, gptAPICall, gptImageCall } from '../../../apis/gpt/GPT';
 import { DocUtils } from '../../../documents/DocUtils';
 import { Docs } from '../../../documents/Documents';
 import { SettingsManager } from '../../../util/SettingsManager';
@@ -145,20 +145,22 @@ export class GPTPopup extends ObservableReactComponent<object> {
                     break;
             } // prettier-ignore
 
-            gptOutput.split('======').filter(item => item.trim() !== '')  // Split output into individual document contents
-                .map(docContentRaw => textToDocMap.get(docContentRaw.replace(/\n/g, ' ').trim())) // the find the corresponding Doc using textToDoc map
-                .filter(doc => doc).map(doc => doc!) // filter out undefined values
-                .forEach((doc, index) => {
+            gptOutput.split(DescriptionSeperator).filter(item => item.trim() !== '')  // Split output into individual document contents
+                .map(docContentRaw => docContentRaw.replace(/\n/g, ' ').trim())
+                .map(docContentRaw => ({doc: textToDocMap.get(docContentRaw.split(DataSeperator)[0]), data: docContentRaw.split(DataSeperator)[1] })) // the find the corresponding Doc using textToDoc map
+                .filter(({doc}) => doc).map(({doc, data}) => ({doc:doc!, data})) // filter out undefined values
+                .forEach(({doc, data}, index) => {
                     switch (questionType) {
                         case GPTDocCommand.Sort:
                             doc[ChatSortField] = index;
                             break;
                         case GPTDocCommand.AssignTags:
-                            if (args) {
-                                const hashTag = args.startsWith('#') ? args : '#' + args[0].toLowerCase() + args.slice(1);
-                                const filterTag = Doc.MyFilterHotKeys.map(key => StrCast(key.toolType)).find(key => key.includes(args)) ?? hashTag;
-                                TagItem.addTagToDoc(doc, filterTag);
-                            }
+                            TagItem.addTagToDoc(doc, data.startsWith('#') ? data : '#'+data);
+                            // if (args) {
+                            //     const hashTag = args.startsWith('#') ? args : '#' + args[0].toLowerCase() + args.slice(1);
+                            //     const filterTag = Doc.MyFilterHotKeys.map(key => StrCast(key.toolType)).find(key => key.includes(args)) ?? hashTag;
+                            //     TagItem.addTagToDoc(doc, filterTag);
+                            // }
                             break;
                         case GPTDocCommand.Filter:
                             TagItem.addTagToDoc(doc, GPTPopup.ChatTag);
@@ -241,7 +243,7 @@ export class GPTPopup extends ObservableReactComponent<object> {
         gptAPICall(userPrompt, GPTCallType.COMMANDTYPE, undefined, true).then((commandType, args = commandType.split(' ').slice(1).join(' ')) =>
             (async () => {
                 switch (this.NumberToCommandType(commandType)) {
-                    case GPTDocCommand.AssignTags:
+                    case GPTDocCommand.AssignTags:return this._documentDescriptions?.then(descs => gptAPICall(userPrompt, GPTCallType.TAGDOCS, descs)) ?? "";
                     case GPTDocCommand.Filter:    return this._documentDescriptions?.then(descs => gptAPICall(userPrompt, GPTCallType.SUBSETDOCS, descs)) ?? "";
                     case GPTDocCommand.Sort:      return this._documentDescriptions?.then(descs => gptAPICall(userPrompt, GPTCallType.SORTDOCS, descs)) ?? "";
                     default:                      return Doc.getDescription(DocumentView.SelectedDocs().lastElement()).then(desc => gptAPICall(userPrompt, GPTCallType.DOCINFO, desc));