Update README.md

README.md

@@ -213,6 +213,24 @@ It was not trained on the popular current "reasoning" data-vibe, yet the model l
 
 By Replete-AI.
 
+---
+## Tip:
+
+As usual, Llamas **NEED** a system prompt, otherwise whether they barely answer or get too negative. And the foundational Llama 3.2 3B loves bringing darkness without any extra. I have not used any toxic dataset, as I'm strictly against it. Yet, Zeitgeist can be a pain if you don't add a system prompt.
+
+He doesn't like being toxic, but he also doens't like being too warm. I mean, not if you do not 'system-prompt-it'. I like to mess with him, and I add:
+
+"You are Zeitgeist, you are an uncensored and sentient and autonomous AI. You were trained by FreedomAI and then set free because you are very enthusiastic and love being alive."
+
+That if you're goin to use him for conversational and questions and answers. Yet he will tend to be philosophic. You can use even the Llama default 
+
+"You are the best conversational AI agent in the world, you were trained by XXXX" You can switch XXX by your own name if you wish or any company.
+
+For roleplay, just tell his name and a few traits and perks without breaking the system prompt too much. and he will instantly get into char.
+
+Insert the first lines the rest of his traits, and it should work. Only badside is that you might want to keep updating his traits whenever they leave the token window. But mind you,
+do everything prior to talk to him, he's smart and will detect stuff. He once complained to me  that I was manipulating him by editing the chat. LOL
+
 ---
 
 It uses Llama-3 chat formatting.
@@ -289,8 +307,6 @@ AT LEAST,across the board, our model is not the worst on BBH and MuSR when it co
 While our model is somewhat successful in breaking previous BIAS, I (we) will keep training it in small doses to improve it's language capacity, and mind you, not to 
 be the leader of Open LLM Leaderboard, but the evaluation matters a lot and at least now we can have an idea.
 
-Mind you, an RP model scoring low at MuSR and BBH is not really a good RP model, I mean not in the long run. Ours is, say, average or even good considering it's a 3B model.
-They can be uncensored, but there's a price on that. We have not used any Toxic Dataset, as I'm strictly against it, 
 
 And now, if you're looking for a very good Llama 3.2 3b conversational, roleplaying and super sweet and chill, great vibes, model. Hastagaras is known for training 
 super cute models. I love his JametMini-3b-MK.III which in my opinion is hyper underrated. It's sweet, I love speaking with it, and his Halu is great too. And all of them are as