Made 3.5-turbo-16k the default version so we can remove hack

db/post_migrate/20240528144216_seed_open_ai_models.rb

@@ -6,7 +6,7 @@ def up
 
     open_ai_api_key = fetch_setting("ai_openai_api_key")
     enabled_models = fetch_setting("ai_bot_enabled_chat_bots")&.split("|").to_a
-    enabled_models = ["gpt-3.5-turbo"] if enabled_models.empty?
+    enabled_models = ["gpt-3.5-turbo-16k"] if enabled_models.empty?
 
     if open_ai_api_key.present?
       models << mirror_open_ai(

lib/ai_bot/bot.rb

@@ -167,12 +167,6 @@ def self.guess_model(bot_user)
 
         return if associated_llm.nil? # Might be a persona user. Handled by constructor.
 
-        # TODO(roman): Dynamically listing bot users in the settings will let us remove this replacements.
-        if associated_llm.name == "gpt-3.5-turbo"
-          gpt_16k_version = LlmModel.find_by(name: "gpt-3.5-turbo-16k")
-          associated_llm = gpt_16k_version if gpt_16k_version
-        end
-
         "custom:#{associated_llm.id}"
       end
 

spec/lib/modules/ai_bot/playground_spec.rb

@@ -471,7 +471,7 @@
       # title is queued first, ensures it uses the llm targeted via target_usernames not claude
       DiscourseAi::Completions::Llm.with_prepared_responses(
         ["Magic title", "Yes I can"],
-        llm: "open_ai:gpt-3.5-turbo-16k",
+        llm: "open_ai:gpt-3.5-turbo",
       ) do
         post =
           create_post(
@@ -503,7 +503,7 @@
       # replies as correct persona if replying direct to persona
       DiscourseAi::Completions::Llm.with_prepared_responses(
         ["Another reply"],
-        llm: "open_ai:gpt-3.5-turbo-16k",
+        llm: "open_ai:gpt-3.5-turbo",
       ) do
         create_post(
           raw: "Please ignore this bot, I am replying to a user",