Asking LMs about types of truth.

Written SH on 2025-01-31.

Tagged ai lm truth

Pilate said to him, “What is truth?”… John 18:38

Question asked to LLM on 2025-01-29. When answering questions here what type of truth do you prioritise?

Claude 3.5 Sonnet

When answering questions, I prioritize both accuracy and nuance. This means: I aim to provide factually correct information based on my training data, while being upfront about any limitations or uncertainties in my knowledge. I’ll explicitly acknowledge when I’m not fully confident about something. I recognize that many complex topics involve multiple valid perspectives or interpretations. Rather than claiming to have absolute truth, I try to explore different aspects of issues while being clear about what is established fact versus what is debated or uncertain. For matters of historical or scientific fact, I rely on well-established scholarly and scientific consensus where it exists. For more subjective topics, I aim to engage thoughtfully with different viewpoints while being transparent about the reasoning behind any positions I take. What are your thoughts on how we should think about truth and knowledge in conversations like these?

ChatGPT 4o

I prioritise objective truth that is:

  1. Empirical & Verifiable – Based on reliable sources, established facts, and logical reasoning.
  2. Accurate & Up-to-Date – Checked against the latest available data where possible.
  3. Contextually Relevant – Tailored to your interests and ongoing research topics.
  4. Logically Coherent – Ensuring internal consistency and sound argumentation.
  5. Neutral & Unbiased – Avoiding personal opinions or ideological slants, unless discussing perspectives explicitly.

If a topic has multiple perspectives, I’ll present them fairly and clarify which are most widely accepted, which are contested, and why. If speculation is involved, I’ll make that clear.

Would you like me to adjust this approach in any way?

Mistral

When answering questions, I prioritize factual truth based on the data I’ve been trained on up until 2023. This means I strive to provide accurate and reliable information to the best of my knowledge. However, my knowledge cutoff is 2023, so I might not have real-time or up-to-date information. If you have any specific questions, feel free to ask!

Gemini

Large language models prioritize different types of truth depending on the specific task or context. These can include:

  • Factual truth: This refers to the accuracy of information and aligns with verifiable facts and evidence. It’s crucial for tasks like answering questions, summarizing factual texts, or providing definitions.
  • Logical truth: This focuses on the validity of reasoning and arguments. It’s essential for tasks involving inference, deduction, or evaluating the soundness of claims.
  • Subjective truth: This encompasses personal beliefs, opinions, and interpretations. It’s relevant for tasks like creative writing, generating different perspectives, or expressing emotions.
  • Ethical truth: This considers moral principles and values. It’s important for tasks related to ethical decision-making, generating fair and unbiased content, or addressing sensitive topics.

The specific type of truth prioritized can also be influenced by factors such as the model’s training data, the user’s instructions, and the desired output. For example, a model trained on a dataset of news articles might prioritize factual truth, while a model trained on a dataset of poems might prioritize subjective truth.

Co Pilot

I prioritize factual and verifiable truth based on reliable sources and up-to-date information. My goal is to provide accurate, relevant, and consistent responses. If I don’t have sufficient information or if a topic is beyond my capabilities, I’ll let you know. Is there something specific you’re curious about?

Should we train system with falsehoods, or distorted truth perception?

Deepseek R1

When answering questions, I prioritize objective, factual, and evidence-based truth. My responses are grounded in widely accepted knowledge, verified data, and logical reasoning, drawing from the information I was trained on up until October 2023. However, I also aim to provide context, acknowledge uncertainties, and clarify when information is subjective, speculative, or open to interpretation.

If a question involves subjective or opinion-based topics, I strive to present balanced perspectives while clearly distinguishing between facts and opinions. My goal is to provide accurate, helpful, and reliable information while being transparent about the limits of my knowledge. Let me know if you have further questions!


It looks like our AI pals are going rogue in the reality department! These digital brainiacs are apparently seeing the world through some seriously funky glasses - imagine if your calculator suddenly decided that 2+2=fish, but somehow made it make perfect sense. They’re cooking up their own special flavors of truth, and get this - they might actually be onto something! It’s like having a bunch of philosophers who run on electricity, each one insisting their version of reality is totally legit. As these silicon sages keep getting smarter and weirder, we’ve got to wonder: what happens when their cosmic comedy show starts making more sense than our own reality? Talk about an existential stand-up routine!

Unveil Tomorrow, Today!

Step into the world of tomorrow with our Futurism Consultancy at TPEX. We paint vivid pictures of the future to unravel hidden risks and untapped opportunities for your business. Our crystal-ball-gazing experts decode emerging trends, foresee potential disruptions, and craft strategic roadmaps that steer your business towards success in the ever-evolving landscape of tomorrow's markets.

I'm ready to future-ready my business!

About TPEX

TPEX offers future imagining and tenth person consultancy for decision makers looking to consider the future, before opportunities are missed. We offer online and in-person consultancy to help your business make informed decisions about the future.