Bug/gpt response compact (#994)

* refactor question builder for gpt feature
* removing logic that prepends all applied tags to question builder
* passing first tag as context to gptservice
* setting context before sending the question

* refactoring setup message

* improving context

* refactoring context for more appropriate responses

* get matching tag or default for context

* sending instructions along with question, instead of setup

* prompt for shorter responses

* values responsible for tweaking AI responses are not declared as constants, docs are also added for such values

Author: ankitsmt211

application/src/main/java/org/togetherjava/tjbot/features/chatgpt/ChatGptCommand.java

@@ -74,8 +74,9 @@ public void onSlashCommand(SlashCommandInteractionEvent event) {
     public void onModalSubmitted(ModalInteractionEvent event, List<String> args) {
         event.deferReply().queue();
 
+        String context = "";
         Optional<String[]> optional =
-                chatGptService.ask(event.getValue(QUESTION_INPUT).getAsString());
+                chatGptService.ask(event.getValue(QUESTION_INPUT).getAsString(), context);
         if (optional.isPresent()) {
             userIdToAskedAtCache.put(event.getMember().getId(), Instant.now());
         }

application/src/main/java/org/togetherjava/tjbot/features/chatgpt/ChatGptService.java

@@ -21,7 +21,30 @@
 public class ChatGptService {
     private static final Logger logger = LoggerFactory.getLogger(ChatGptService.class);
     private static final Duration TIMEOUT = Duration.ofSeconds(90);
+
+    /** The maximum number of tokens allowed for the generated answer. */
     private static final int MAX_TOKENS = 3_000;
+
+    /**
+     * This parameter reduces the likelihood of the AI repeating itself. A higher frequency penalty
+     * makes the model less likely to repeat the same lines verbatim. It helps in generating more
+     * diverse and varied responses.
+     */
+    private static final double FREQUENCY_PENALTY = 0.5;
+
+    /**
+     * This parameter controls the randomness of the AI's responses. A higher temperature results in
+     * more varied, unpredictable, and creative responses. Conversely, a lower temperature makes the
+     * model's responses more deterministic and conservative.
+     */
+    private static final double TEMPERATURE = 0.8;
+
+    /**
+     * n: This parameter specifies the number of responses to generate for each prompt. If n is more
+     * than 1, the AI will generate multiple different responses to the same prompt, each one being
+     * a separate iteration based on the input.
+     */
+    private static final int MAX_NUMBER_OF_RESPONSES = 1;
     private static final String AI_MODEL = "gpt-3.5-turbo";
 
     private boolean isDisabled = false;
@@ -42,18 +65,16 @@ public ChatGptService(Config config) {
 
         openAiService = new OpenAiService(apiKey, TIMEOUT);
 
-        ChatMessage setupMessage = new ChatMessage(ChatMessageRole.SYSTEM.value(),
-                """
-                        Please answer questions in 1500 characters or less. Remember to count spaces in the
-                        character limit. For code supplied for review, refer to the old code supplied rather than
-                        rewriting the code. Don't supply a corrected version of the code.\s""");
+        ChatMessage setupMessage = new ChatMessage(ChatMessageRole.SYSTEM.value(), """
+                For code supplied for review, refer to the old code supplied rather than
+                rewriting the code. DON'T supply a corrected version of the code.\s""");
         ChatCompletionRequest systemSetupRequest = ChatCompletionRequest.builder()
             .model(AI_MODEL)
             .messages(List.of(setupMessage))
-            .frequencyPenalty(0.5)
-            .temperature(0.3)
+            .frequencyPenalty(FREQUENCY_PENALTY)
+            .temperature(TEMPERATURE)
             .maxTokens(50)
-            .n(1)
+            .n(MAX_NUMBER_OF_RESPONSES)
             .build();
 
         // Sending the system setup message to ChatGPT.
@@ -68,26 +89,29 @@ public ChatGptService(Config config) {
      * @see <a href="https://platform.openai.com/docs/guides/chat/managing-tokens">ChatGPT
      *      Tokens</a>.
      */
-    public Optional<String[]> ask(String question) {
+    public Optional<String[]> ask(String question, String context) {
         if (isDisabled) {
             return Optional.empty();
         }
 
         try {
-            ChatMessage chatMessage =
-                    new ChatMessage(ChatMessageRole.USER.value(), Objects.requireNonNull(question));
+            String instructions = "KEEP IT CONCISE, NOT MORE THAN 280 WORDS";
+            String questionWithContext = "context: Category %s on a Java Q&A discord server. %s %s"
+                .formatted(context, instructions, question);
+            ChatMessage chatMessage = new ChatMessage(ChatMessageRole.USER.value(),
+                    Objects.requireNonNull(questionWithContext));
             ChatCompletionRequest chatCompletionRequest = ChatCompletionRequest.builder()
                 .model(AI_MODEL)
                 .messages(List.of(chatMessage))
-                .frequencyPenalty(0.5)
-                .temperature(0.3)
+                .frequencyPenalty(FREQUENCY_PENALTY)
+                .temperature(TEMPERATURE)
                 .maxTokens(MAX_TOKENS)
-                .n(1)
+                .n(MAX_NUMBER_OF_RESPONSES)
                 .build();
 
             String response = openAiService.createChatCompletion(chatCompletionRequest)
                 .getChoices()
-                .get(0)
+                .getFirst()
                 .getMessage()
                 .getContent();
 

application/src/main/java/org/togetherjava/tjbot/features/help/HelpSystemHelper.java

@@ -125,7 +125,11 @@ RestAction<Message> constructChatGptAttempt(ThreadChannel threadChannel,
         String question = questionOptional.get();
         logger.debug("The final question sent to chatGPT: {}", question);
 
-        chatGPTAnswer = chatGptService.ask(question);
+        ForumTag defaultTag = threadChannel.getAppliedTags().getFirst();
+        ForumTag matchingTag = getCategoryTagOfChannel(threadChannel).orElse(defaultTag);
+
+        String context = matchingTag.getName();
+        chatGPTAnswer = chatGptService.ask(question, context);
         if (chatGPTAnswer.isEmpty()) {
             return useChatGptFallbackMessage(threadChannel);
         }
@@ -178,20 +182,6 @@ private Optional<String> prepareChatGptQuestion(ThreadChannel threadChannel,
             .min(MAX_QUESTION_LENGTH - questionBuilder.length(), originalQuestion.length()));
 
         questionBuilder.append(originalQuestion);
-
-        StringBuilder tagBuilder = new StringBuilder();
-        int stringLength = questionBuilder.length();
-        for (ForumTag tag : threadChannel.getAppliedTags()) {
-            String tagName = tag.getName();
-            stringLength += tagName.length();
-            if (stringLength > MAX_QUESTION_LENGTH) {
-                break;
-            }
-            tagBuilder.append(String.format("%s ", tagName));
-        }
-
-        questionBuilder.insert(0, tagBuilder);
-
         return Optional.of(questionBuilder.toString());
     }