LLM Prompt settings

Taming the Local LLM: How System Prompts Unlock "Veritas"

Date: November 28, 2025

Model: soob3123_Veritas-12B-Q8_0.gguf 

Today’s mission was to take a local LLM off the leash. I’ve been setting up Veritas-12B, a high-precision philosophical model, running locally via Ollama and AnythingLLM3.

The hardware is capable—a Framework Desktop with 128GB of RAM—so I knew the bottleneck wasn't the machine. It was the instructions. The difference between a generic "helpful assistant" and a true "research partner" lies entirely in how you prompt the system before you ask the first question.
Here is the log of how we moved from a generic summary to a post-doctoral deep dive.

Phase 1: The Generic "Assistant"

Initially, AnythingLLM was set to its default system prompt. I asked a standard philosophical question about Jean-Paul Sartre.

System Prompt: “You are a helpful AI assistant” 
My Prompt: “Tell me what Satre says about deceiving oneself” 

The Response:
“Sartre argues that self-deception is a fundamental aspect of human existence and a means by which individuals attempt to cope with the anxieties and uncertainties of life. He suggests that self-deception can provide a temporary escape from the burden of freedom and responsibility, allowing individuals to create a narrative that provides a sense of coherence and meaning in their lives. However, Sartre emphasizes that this self-deception ultimately leads to inauthenticity and a denial of one’s true freedom.” 

Analysis:
It was accurate, but shallow. It took about 29 seconds to generate7. It felt like a Wikipedia summary—safe, concise, and uninspired.

Phase 2: The Persona Shift

I decided to push the model by manually injecting a persona into the chat. I wanted depth, so I gave it a specific role to play.

New Prompt: “You are a post doctoral student in the history of philosophy. Answer the user's questions comprehensively and in great detail. elaborate on philosophical concepts and provide extensive examples.” 

The Response:
“Jean-Paul Sartre, a prominent figure in existentialist philosophy, discusses self-deception in his work, particularly in "Being and Nothingness." According to Sartre, self-deception occurs when an individual chooses to deny or ignore aspects of their own existence or the truth of their situation, often as a way to avoid responsibility or anxiety.
Sartre emphasizes that self-deception is a form of "bad faith," where individuals attempt to escape their freedom and the responsibility that comes with it. He argues that by deceiving oneself, one is essentially denying one's own existence and authenticity. This act of self-deception can manifest in various forms, such as rationalizing one's choices, avoiding difficult emotions, or pretending to be something one is not.
Sartre believes that recognizing self-deception is crucial for living authentically. He encourages individuals to confront their true nature and the implications of their choices, even if it leads to discomfort or anxiety. Embracing one's freedom and responsibility, rather than hiding from them, is central to Sartre's philosophy.” 

Analysis:
This was better. It introduced the specific term "bad faith" and expanded on the concepts. However, it was still a wall of text—a single stream of thought without structure.

Phase 3: Unlocking the Architecture

The real breakthrough happened when I realized I wasn't just fighting the prompt; I was fighting the interface settings. I navigated to the Workspace Settings > Chat Settings in AnythingLLM and replaced the default system prompt globally to "Unlock Model Verbosity".

By hard-coding the "Post-Doctoral" instruction into the system layer (rather than just the chat window), the model fully engaged its formatting capabilities.

The "Unlocked" Response:

“Jean-Paul Sartre, a prominent figure in existentialist philosophy, addresses the concept of self-deception, or "mauvaise foi," extensively in his works, particularly in "Being and Nothingness" (1943) and "Existentialism Is a Humanism" (1946). Sartre's exploration of self-deception revolves around the idea that humans often try to escape the anxiety and responsibility that come with their freedom by deceiving themselves about their own nature and choices.

The Nature of Self-Deception (Mauvaise Foi)

Self-deception occurs when an individual attempts to deny or ignore their own freedom and the implications of that freedom. Sartre argues that people often seek comfort in pre-defined roles or identities, which can be seen as a form of bad faith. In this way, they attempt to alleviate the burden of existential angst by adopting a fixed identity rather than embracing the fluidity and uncertainty of existence.
For Sartre, the concept of "mauvaise foi" arises from the inherent contradiction in human existence: we are free beings, yet we often seek to escape this freedom by conforming to external expectations or social roles. This contradiction leads to a form of self-deception where individuals fail to recognize their own freedom and instead act as if they are determined by external factors, such as societal norms or prescribed identities.

Examples of Self-Deception

The Waiter Example: Sartre famously illustrates self-deception with the example of a waiter who performs his role with exaggerated movements and a forced smile. The waiter embodies the role of a waiter to the point where he loses sight of his own identity. He does this to escape the discomfort of being himself and to conform to the expectations of his role. This example shows how individuals can deceive themselves into believing that they are defined by their profession rather than recognizing their freedom to choose.
The Bourgeois Woman: Sartre describes a bourgeois woman who identifies strongly with her social class and the associated values and expectations. She sees herself as a woman of a certain type, defined by her possessions and lifestyle. By clinging to this identity, she avoids confronting the existential questions about her freedom and the choices she could make, thus deceiving herself about her true nature.
The Role of Society: Sartre emphasizes that society plays a significant role in shaping self-deception. People often adopt societal roles that are imposed upon them, such as gender roles or class expectations. For instance, a woman may feel pressured to conform to traditional notions of femininity, leading her to deny her own desires and ambitions. This conformity can lead to a form of self-deception where she believes she is fulfilling her true self when, in reality, she is merely playing a role dictated by societal norms.

Consequences of Self-Deception

Self-deception can have profound consequences for an individual's life. Sartre argues that when people deceive themselves, they fail to engage authentically with their own existence and the choices they face. This can lead to a sense of alienation, where individuals feel disconnected from their true selves and from others.
Moreover, self-deception can stifle personal growth and development. By denying their freedom and the responsibility that comes with it, individuals may miss opportunities for self-discovery and authentic engagement with the world. This can lead to a life characterized by inauthenticity and a lack of fulfillment.

Conclusion

In summary, Jean-Paul Sartre's exploration of self-deception highlights the complexities of human existence and the challenges of embracing one's freedom. Self-deception, or "mauvaise foi," is a pervasive aspect of human life, often arising from the tension between our freedom and the expectations of society. By recognizing and confronting this tendency, individuals can move toward a more authentic and fulfilling existence, embracing their responsibility and the myriad possibilities that come with being free.” 

The Lesson

The difference between the first and last response is not the model—it's the system configuration. The Q8 quantization of Veritas-12B has the knowledge12, but without the correct system prompt settings in AnythingLLM, it was holding back. Once unlocked, it provided structure, specific examples (the famous "Waiter" and "Bourgeois Woman"), and distinct formatting.

We also solved a "Retrieval Failure" regarding a Philosophy PDF by adjusting the chunk size from ~1400 tokens down to 51213131313. This proves that running local AI is as much about the setup as it is about the raw compute power.