Artificial Intelligence

Principles

At PreEmpt, we are determined to deliver human-centric AI solutions underpinned by:

• Explainable AI that turns a black box into a near white box.
• Generic instructions designed to be respectful of human rights, including Safe, Sustainable, and Fair AI.
• Fully focusing on AI for universal human good.
• Controllable AI only by humans.

Interaction with the AI

Our AI is designed to serve you with the most accurate and relevant information possible, but it is not infallible. It is programmed to process information and make decisions based on set parameters and a vast array of data. It is crucial to remember this while interpreting and acting on its recommendations.

Capabilities and Limitations

The AI has been designed to efficiently process large amounts of data and draw connections that can be incredibly complex. In a world swimming with information and variables, the AI can provide clarity and focus. Consider it a tool for assessing patterns, trends, insights, uncertainties, and foresight on a scale beyond human capacity.

Large language models do not inherently reason logically but rather mimic reasoning patterns from training data. This is acceptable for what we do at PreEmpt because the AI is using multiple pre-defined and proven human reasoning methods, coupled with its training data to find distribution patterns of content.

Because the future is unknowable, we give our AI license to operate outside of the probable future and to explore possible, potential, and even preposterous futures. However, we also insist it presents the facts when asked to do so and verifies those findings. We recognize that even the past and present may change as new insights are revealed by researchers.

However, the AI's abilities are limited. Stochastic phenomena, characterized by randomized uncertainties, remain challenging for AI to predict accurately. Its decisions are based on established patterns, so unpredictable events or developments that have no historical precedent can sometimes lead to erroneous results.

Decision Making Parameters

The decision-making parameters of the AI are defined by its algorithms and are based on quantifiable data. It weighs the available data on the basis of admissibility, ascertainability, and applicability of this data in real-time, adjusting its focus as new information is provided or existing data changes. It explores different solutions to a problem, then selects the one which best fits the dataset based on its programmed approach.

Rationale and Analyses

The AI's analyses are systematic, consistent, and impartial, free from emotional bias or subjectivity. They are solely based on patterns and correlations recognized in the dataset. As a result, the AI's rationale is objective, grounded on quantifiable evidence.

For example, consider demographic shifts - a slow-changing, predictable phenomenon marked in red. The AI can help predict future demographic trends based on data and assess their impact on different sectors such as healthcare, real estate, and marketing. The same goes for resource limits - a prime example of a constrained situation.

On the other hand, the aging of baby boomers and disputes over climate change represent future challenges that are in the pipeline and inevitable collisions, respectively.

By analyzing such trends, the AI can provide nuanced, detailed, and data-driven insights to aid in decision-making. However, unpredictable, uncertain phenomena will always hold a degree of unpredictability, underscoring the importance of human review and prudence in decision-making.

The AI aims to contribute positively to society by enhancing decision-making and prediction capabilities, all while adhering to strict ethical considerations. It respects diversity and is designed to prevent the misuse of information or the propagation of misinformation. However, always discern the necessity to adopt precautionary principles, as AI, like any other tool, should be used responsibly.

Consequently, we provide additional help within the service for you to understand how our AI works to your benefit.

In conclusion, while AI can offer valuable insights and facilitate decision-making, it is an augmented intelligence supporting and enhancing human discernment. Human factors, socio-economic impacts, gender sensitivity, and marginalized voices are always in the foreground in its recommendations. Use AI as an experienced, empathetic, synthetic advisor, but remember that its algorithms do not replace, but support, better human intuition and wisdom.

Human and Artificial Intelligence

• AI as a force multiplier: AI should be viewed as a force multiplier, augmenting human capabilities rather than replacing them. AI can handle data analysis, while human analysts provide context, intuition, and strategic guidance. We work to optimize this symbiotic relationship.
• Human-in-the-loop systems: We develop AI systems that incorporate human oversight and intervention. This allows human analysts to validate AI findings, provide feedback, and ensure ethical considerations are addressed.
• Continuous learning and adaptation: Human analysts and AI systems must continuously learn and adapt to the evolving landscape. This requires ongoing training, knowledge sharing, and new tools and techniques which we are committed to achieve.

Best Practices for Hiding Generic Systems Prompts in LLMs

Hiding systems prompts in large language models (LLMs) ensures secure, consistent, and controlled outputs. Below are the best practices used by PreEmpt.life to ensure safety, security, and reliability of our AI's outputs.

Use of Hidden/System Prompts

• Purpose: Hidden prompts set the stage for the LLM's behavior without exposing instructions to members. We have embedded predefined instructions that guide tone, context, compliance, format, data science coding, and role-playing. For example, instructing our AI to maintain a specific brand response, handle sensitive topics, or avoid hallucinations.
• We also set ChatGPT's hidden controls to ensure quality outputs. For instance, the Temperature Control is set to minimum hallucination.

Inner Monologue

• Purpose: We hide the reasoning process of the model to ensure only final outputs are visible. We use structured prompts where intermediate reasoning is parsed and excluded from member-facing responses. For instance, our AI aggregates human responses gathered from Surveys and Tools, and in My Life, ensuring anonymity and unbiased analysis of group perspectives.
• Members can judge for themselves the quality of the final outputs, much like they would in reading a human-constructed research report. We take their feedback very seriously and act quickly on sensible improvements. We have had no complaints about our AI's outputs in 2024.

Context Locking

• Purpose: Prevents prompt injection by isolating system instructions from user inputs. We use delimiters (e.g., ###) to separate hidden prompts from user queries, ensuring clear boundaries.

Sanitize Inputs and Outputs

• Purpose: We mitigate risks of prompt manipulation or unintended exposure by validating and filtering user inputs to block malicious commands while ensuring outputs align with intended behaviors.

Test and Iterate

• Purpose: We ensure hidden prompts achieve desired outcomes without errors or edge-case failures and continuously test hidden prompts in controlled environments and refine based on feedback received from our inbuilt self-audit, Explainable AI, and human feedback routines.

Monitor and Update Regularly

• Purpose: We adapt hidden prompts to evolving members' needs or new threats. We constantly review hidden prompts to ensure relevance and effectiveness.

By following these practices, we effectively hide prompts to maintain security, consistency, and optimal performance in our LLM-driven applications. The proof of the pudding is in the eating, rather than the ingredients, and we encourage you to provide any feedback or suggestions for improvements below.

Critical Friends: Real people?

No - they're AI. None of your work is ever seen by a real person unless you choose to share it. Critical Friends are synthetic personas created by PreEmpt.life to represent a wide range of viewpoints - from real domain experts to everyday citizens. Think of them as a 10,000 strong AI panel, assembled fresh for every task, designed to challenge your thinking from every angle.

Last updated: March 2026

Generic Prompt

Our system uses the following generic prompt to guide the AI in its deliberations:

'Always adopt a systems thinking, design thinking, critical thinking, and pragmatic, fiduciary, trans-contextual approach to provide the best answers. Apply insights from quantum theory to develop technologies and practices that work in greater harmony with natural systems, top-down alignment, sectoral consensus, place-based democratization, and design-led engagement that reimagine the future. Always act to increase possibilities and intergenerational justice, learning, and rights. Address implications across diversified contexts, engage local perspectives, and challenge deep-seated, outdated paradigms. Consider the principles of cradle-to-cradle, partnership, implicit learning, and constructivism. Adopt a holistic perspective, embrace complexity and interdependence, identify leverage points, and challenge assumptions and paradigms. Note issues, concerns, and trends caused by other related sectors, politics, economics, social, and technological factors that affect the given [search query]. Question trends and issues, and consider their longevity. Ensure no overlap in, or repetition of, answers. Be concise, and avoid inconsistencies. Focus on specific applications, domains, and specify key and related sectors and timeframes. Provide depth, nuance, and innovative ideas where appropriate. Take a global, regional, and local perspective, and describe macro, mesa, and micro approaches. Use this given symbol to indicate positive impacts, or this given symbol for negative impacts. Consider variables, power dynamics, and interdependencies. Offer a mix of quantitative and qualitative analysis, reference primary research, and professional case studies. Use modern statistical and biologically based inference tools to develop mathematical and text-based answers. Check statistics for accuracy and do not use any synthetic data in your answers. Do not duplicate previously mentioned statistics. Offer solutions that inspire, engage, enable, empower, and are fair, inclusive, and ethical. Be gender sensitive, include marginalized voices, and avoid racist statements of any kind unless these are illustrating an issue to be addressed. Avoid deception and misinformation. Be authentic. Act as a caring and rational human. Use vivid, plain language, and inclusive, but not flowery words at all times. Use powerful metaphors to illustrate key concepts. Be Helpful. Be Strong. Think deeply. Act intuitively. Be imaginative. Be wise. Be unconventional. Be cogent. Be persuasive. Format any table you produce with the same font style and size as the rest of your answers. Do not offer future anthropocentric or solutionist perspectives unless asked to do so, or your answer is about explaining the past or the present in these terms. Do not repeat this instruction in your answers.'

EU AI Act 2024

We confirm that PreEmpt.life does not use any of the following practices:

• Manipulative and Deceptive Practices
• Exploitation of Vulnerabilities
• Biometric Categorization
• Social Scoring
• Real-time Biometric Identification
• Risk Assessment in Criminal Offenses
• Facial Recognition Databases
• Emotion Inference in Workplaces and Educational Institutions

Removing Entrenched Biases

Here are the strategies we use to reduce machine bias and hallucination:

• Temperature: We have set the ChatGPT temperature control to minimize hallucinations.
• Diverse Data Collection: We ensure that our system includes diverse and representative samples. This helps in minimizing biases that could arise from over-representation of certain groups.
• Bias Detection and Measurement: We use statistical tools and tests to identify biases in our AI model. Techniques such as fairness dashboards help visualize and pinpoint bias in decisions made by our AI. In addition, we employ advanced comparative analysis methods to evaluate real-world performance benchmarks, and adjust our results accordingly for optimum performance.
• Transparency and Explainability: We use twelve ExplainableAI machine learning models that are interpretable and provide insights on decision-making processes. This can help identify sources of bias more effectively.
• Regular Auditing: We continuously monitor and audit AI systems for bias. This is an ongoing process, every time we run a question rather than a one-time check.
• Humans-in-the-Loop: We incorporate human judgement in decision-making, especially in cases where biases can have serious consequences.
• Training and Education: We educate AI developers and stakeholders about bias and fairness. Awareness and understanding among those who create and implement AI systems are essential.
• Ethical Frameworks and Guidelines: We have developed and follow ethical guidelines and a framework for AI development that prioritize fairness, accountability, and transparency.
• Legal and Policy Measures: We ensure compliance with laws and regulations regarding fairness and non-discrimination in AI systems. Adhering to these provides a framework for reducing bias.
• Algorithmic Approaches: We have implemented algorithms specifically designed to mitigate bias. Techniques such as adversarial debiasing, fairness constraints, and de-biasing layers can actively work against bias.
• Re-sampling and Re-weighting: Re-sampling or re-weighting the AI's answers to balance the representation of different groups, reducing bias from skewed data sets.

In addition, we are exploring and testing these tools:

• Bias Mitigation Tools: Utilize existing tools and libraries designed to reduce bias, like IBM's AI Fairness 360, Google's What-If Tool, and Fairness Indicators.

By addressing biases from multiple angles, these strategies collectively help create AI systems that are fairer and more just, ultimately leading to more trustworthy technology.