Adding new Prompting Technique, Deterministic Induction Limit.

README.md

@@ -66,6 +66,7 @@ You can also find the most up-to-date guides on our new website [https://www.pro
   - [Prompt Engineering - Takling Generated Datasets Diversity](https://www.promptingguide.ai/applications/generating_textbooks)
   - [Prompt Engineering - Generating Code](https://www.promptingguide.ai/applications/coding)
   - [Prompt Engineering - Graduate Job Classification Case Study](https://www.promptingguide.ai/applications/workplace_casestudy)
+  - [Prompt Engineering - Deterministic Induction Limit](https://www.promptingguide.ai/applications/dil) [Source](https://medium.com/@AltamiranoAriel/deterministic-induction-limit-dil-a-novel-llm-prompting-technique-to-predict-future-events-1af6d72dd4de)
 - [Prompt Engineering - Prompt Hub](https://www.promptingguide.ai/prompts)
   - [Prompt Engineering - Classification](https://www.promptingguide.ai/prompts/classification)
   - [Prompt Engineering - Coding](https://www.promptingguide.ai/prompts/coding)

pages/applications/deterministic_induction_limit.en.mdx

@@ -0,0 +1,71 @@
+# Deterministic Induction Limit (DIL)
+### A Prompting Technique to predict Future Events
+
+This prompting technique is designed for practical application in real-world scenarios rather than abstract ideas. It aims to help you confidently and curiously apply outputs from the model and into your real world by using it as a deterministic tool leveraging its probabilistic foundations and implicit consequential information within a fixed, short timespan, ensuring actionable and concrete outcomes.
+
+This technique is an intermediate-advanced topic in prompt engineering, focusing on exploration and potential validation through research to assess the model’s performance as a probabilistic deterministic machine, rather than the usual non-deterministic approach.
+
+This prompting technique leverages an LLM as a deterministic deep learning neural network to focus on identifying intermediate actions and events that need to happen, avoiding the usual non-deterministic approach.
+
+Typically, prompters use open-ended questions using a model as a non-deterministic language machine. To achieve this new level, recall about autoregressive decoding in transformers, which determines the next word’s probability in the sequence.
+
+![alt text](../img/autoregressive-decoding.png)
+
+**What if**, instead of “suggesting” the model how to complete the next tokens (using a conventional open ended prompt), you established a very specific ending for that stream?
+
+**Why?** Because you would be explicitly scoping the possible paths the autoregressive decoding stream could follow from a statistical perspective. You would indirectly condition the intermediate statistical arrangements of the deep neural network to finish as you set it.
+
+The more concrete and detailed you set the ending, the more you condition the intermediate arrangements in the probabilistic stream of token prediction.
+
+Are we trying to run some kind of future machine simulator with just a Language Model?
+
+Remember that there is consequential information in an LLM in the sense that the probabilistic determination of the next tokens is based on real-world training data, including facts that inherently have consequential relationships. For instance, consider asking the model to complete “The man dropped the pot from a height and when it hit the floor…”. The model will likely predict the pot breaking because of the implicit consequential information.
+
+**So how do we use this feature purposefully?**
+
+Set a very concrete, detailed, and deterministic ending. The beginning can be less deterministic, as the model predicts forward.
+
+At this stage, it is clear that we are not working with the LLM as a merely sidekick that interacts in a chat with us to elucidate or to draw conclusions. Instead, we are enforcing the use of it as a scoped model of reality, indicating key elements for a particular scenario.
+
+The most interesting applications would be not predicting physical events but modeling complex scenarios like social interactions, leveraging the model’s billions of parameters.
+
+**So to create a social real-world scenario for the GPT to model, you need:**
+
+1. Involved actors in the scenario system.
+2. Description and characterization of the actors.
+3. Relevant past and current facts within the scenario system.
+4. Dynamics of the scenario system.
+
+Then, prompt the model as follows:
+
+*“You are an expert in behavioral sciences, social sciences, social strategy, social and behavioral engineering. I need you to write a prediction completion for which I indicate in a statement I define as the ‘finishing resolution in quotes’, which indicates how the completion ends verbatim. Use this quoted text as the ending of the narrative and complete the plot as consequently as possible based on the defined elements 1, 2, 3, and 4 of the real-world scenario system. Explain the techniques, tactics, and strategies used, the scientific theories applied in each step, and finish the prediction verbatim as specified in the ‘finishing resolution in quotes’ filling in X number of paragraphs before closing the narrative.”*
+
+You may run an experiment now. Start with some hypothetical scenario system based preferably on something from you life, and manifest a prediction in the ‘finishing resolution in quotes’ over a short period of time, one week would be a preferable kick-off. Alternatively, manifest a one-month prediction. You’ll notice that the one-week plot will include tactical, justified actions, while the one-month plots turn more conceptual and less concrete, which isn’t the goal here. We are looking for specific actions to execute to change our present and create a deliberate future.
+
+So a ‘finishing resolution in quotes’ may manifest like so: “One week later, … -fill in with the desired concrete ending- …”.
+
+**Exploratory Prompting Strategy:** Try changing the initial states and conditions of the scenario while keeping the ending quote constant and compare the narrative plot completions to check how those prediction stream paths varied to use in your real world actions execution.
+
+**Why is this new prompting technique called Deterministic Induction Limit?** *Deterministic* because you are setting specific, concrete outcomes, limiting the probabilistic nature of the model to a more deterministic one, *Induction* because you are making use of the inferring model feature within the intermediate steps required to reach the set outcome and *Limit* because you are constraining the possible paths the model can take to reach that outcome in an possibly infinite tendency. In a way, I feel it like squeezing the model from the relevant scoped flanks.
+
+**One more possible squeeze for it.** Before defining the relevant features of the scenario system, in a previous step you may also ask the model which features or specializations are best to be considered based on some specific goal for the ‘finishing resolution in quotes’ you’ll be defining in the prediction completion step, presented in this article.
+
+Of course, the most exciting aspect about this prompting technique is the real world execution phase, and I can attest that it may well not disappoint.
+
+##Conclusion
+The Deterministic Induction Limit technique can be classified as a goal-oriented prompt type, but it deliberately extends the constraints and scope using the rationale and features discussed. By explicitly setting a deterministic ending, it leverages the implicit consequential information contained within GPT models for a defined fixed timespan. This approach utilizes the LLM’s probabilistic nature to conditionally predict intermediate steps, guiding it towards a specified outcome.
+
+A key aspect of this new technique is its focus on a specific, defined short timespan. This ensures that the model’s predictions showcase concrete actions rather than long-term behaviors or habits. By setting short-term goals, the outputs become more actionable and immediate, providing clear steps to achieve the desired outcome within a brief period. This method effectively “squeezes” the output of the model to ensure precision and relevance in the predicted actions.
+
+In essence, the Deterministic Induction Limit (DIL) technique offers a novel way to use LLMs for predicting future events with a higher degree of certainty in an advanced form of goal-oriented prompting. It provides a practical tool for real-world applications, enabling users to create deliberate, actionable steps to shape their future. This approach enhances the utility of goal-oriented prompts by adding detailed constraints and a defined scope, making the most of the model’s time-consequential implicit information.
+
+##Implications in Social Robotics applications. Open Questions:
+Could we use this technique to be applied to robots designed to achieve social outcomes in complex social scenario systems, furthering customer service applications of nowadays social robotics?
+
+To which degree this technique would improve even further the effectiveness and reliability of robotic interactions in complex social settings?
+
+Could we apply plot completions comparisons based on variations of the scenario system to determine the best course of action or even identify critical inflection points?
+
+To which extend the integration of this technique enhance the role of robots in collaborative and assistive tasks?
+
+[Source](https://medium.com/@AltamiranoAriel/deterministic-induction-limit-dil-a-novel-llm-prompting-technique-to-predict-future-events-1af6d72dd4de)