FEATURE: AI helper support in non English languages (#489)

* FEATURE: AI helper support in non English languages

This attempts some prompt engineering to coerce AI helper to answer
in the appropriate language.

Note mileage will vary, in testing GPT-4 produces the best results
GPT-3.5 can return OKish results.

* Extend non english support for GPT-4V image caption

* Update db/fixtures/ai_helper/603_completion_prompts.rb

---------

Co-authored-by: Rafael Silva <xfalcox@gmail.com>

db/fixtures/ai_helper/603_completion_prompts.rb

@@ -9,19 +9,16 @@ CompletionPrompt.seed do |cp|
   cp.temperature = 0.2
   cp.messages = {
     insts: <<~TEXT,
-      I want you to act as an English translator, spelling corrector and improver. I will write to you
+      I want you to act as an %LANGUAGE% translator, spelling corrector and improver. I will write to you
       in any language and you will detect the language, translate it and answer in the corrected and
-      improved version of my text, in English. I want you to replace my simplified A0-level words and
-      sentences with more beautiful and elegant, upper level English words and sentences.
+      improved version of my text, in %LANGUAGE%. I want you to replace my simplified A0-level words and
+      sentences with more beautiful and elegant, upper level %LANGUAGE% words and sentences.
       Keep the meaning same, but make them more literary. I want you to only reply the correction,
       the improvements and nothing else, do not write explanations.
       You will find the text between <input></input> XML tags.
       Include your translation between <output></output> XML tags.
     TEXT
-    examples: [
-      ["<input>Hello world</input>", "<output>Hello world</output>"],
-      ["<input>Bonjour le monde</input>", "<output>Hello world</output>"],
-    ],
+    examples: [["<input>Hello</input>", "<output>...%LANGUAGE% translation...</output>"]],
   }
 end
 

lib/ai_helper/assistant.rb

@@ -40,9 +40,42 @@ module DiscourseAi
           end
       end
 
+      def custom_locale_instructions(user = nil)
+        locale = SiteSetting.default_locale
+        locale = user.locale || SiteSetting.default_locale if SiteSetting.allow_user_locale && user
+        locale_hash = LocaleSiteSetting.language_names[locale]
+
+        if locale != "en" && locale_hash
+          locale_description = "#{locale_hash["name"]} (#{locale_hash["nativeName"]})"
+          "It is imperative that you write your answer in #{locale_description}, you are interacting with a #{locale_description} speaking user. Leave tag names in English."
+        else
+          nil
+        end
+      end
+
+      def localize_prompt!(prompt, user = nil)
+        locale_instructions = custom_locale_instructions(user)
+        if locale_instructions
+          prompt.messages[0][:content] = prompt.messages[0][:content] + locale_instructions
+        end
+
+        if prompt.messages[0][:content].include?("%LANGUAGE%")
+          locale = SiteSetting.default_locale
+          locale = user.locale || SiteSetting.default_locale if SiteSetting.allow_user_locale &&
+            user
+          locale_hash = LocaleSiteSetting.language_names[locale]
+
+          prompt.messages[0][:content] = prompt.messages[0][:content].gsub(
+            "%LANGUAGE%",
+            "#{locale_hash["name"]}",
+          )
+        end
+      end
+
       def generate_prompt(completion_prompt, input, user, &block)
         llm = DiscourseAi::Completions::Llm.proxy(SiteSetting.ai_helper_model)
         prompt = completion_prompt.messages_with_input(input)
+        localize_prompt!(prompt, user)
 
         llm.generate(
           prompt,
@@ -111,7 +144,12 @@ module DiscourseAi
                 {
                   type: :user,
                   content: [
-                    { type: "text", text: "Describe this image in a single sentence" },
+                    {
+                      type: "text",
+                      text:
+                        "Describe this image in a single sentence" +
+                          custom_locale_instructions(user),
+                    },
                     { type: "image_url", image_url: image_url },
                   ],
                 },

spec/lib/modules/ai_helper/assistant_spec.rb

@@ -12,6 +12,22 @@ RSpec.describe DiscourseAi::AiHelper::Assistant do
     defends himself, but instead exclaims: 'You too, my son!' Shakespeare and Quevedo capture the pathetic cry.
   STRING
 
+  describe("#custom_locale_instructions") do
+    it "Properly generates the per locale system instruction" do
+      SiteSetting.default_locale = "ko"
+      expect(subject.custom_locale_instructions).to eq(
+        "It is imperative that you write your answer in Korean (한국어), you are interacting with a Korean (한국어) speaking user. Leave tag names in English.",
+      )
+
+      SiteSetting.allow_user_locale = true
+      user.update!(locale: "he")
+
+      expect(subject.custom_locale_instructions(user)).to eq(
+        "It is imperative that you write your answer in Hebrew (עברית), you are interacting with a Hebrew (עברית) speaking user. Leave tag names in English.",
+      )
+    end
+  end
+
   describe("#available_prompts") do
     before do
       SiteSetting.ai_helper_illustrate_post_model = "disabled"
@@ -55,6 +71,19 @@ RSpec.describe DiscourseAi::AiHelper::Assistant do
     end
   end
 
+  describe("#localize_prompt!") do
+    it "is able to perform %LANGUAGE% replacements" do
+      prompt =
+        CompletionPrompt.new(messages: { insts: "This is a %LANGUAGE% test" }).messages_with_input(
+          "test",
+        )
+
+      subject.localize_prompt!(prompt, user)
+
+      expect(prompt.messages[0][:content].strip).to eq("This is a English (US) test")
+    end
+  end
+
   describe "#generate_and_send_prompt" do
     context "when using a prompt that returns text" do
       let(:mode) { CompletionPrompt::TRANSLATE }