AI Policy

Last updated June 10, 2026

Our projects leverage AI heavily, however in quite a specific way. This page serves to clarify our conception of it and how we use it. It is quite in depth, however this subject merits it.

What an LLM actually is

LLMs in essence are complex algorithms for determining the most probable text to complete the input text.

You have some input text such as "What is the weather in Portland, Oregon". You pass this into an LLM and it receives "What is the weather in Portland, Oregon" along with some preceding instructions about what to do. That's called the system prompt.

The LLM receives this as one block of text, it does not have any notion of what a system or a user is, it is simply an algorithm, essentially a mathematical function.

It takes that input text block, system and user, and converts it into a giant set of numbers. User and system are not distinguished in this set of numbers as they are in the text, it's essentially a long list of numbers indifferent to the content or sections you think you have included in the text.

Given that set of numbers, it attempts to use some mathematical formulas and structures it has embedded in it to try and determine what is the most likely next piece of text (or more specifically its mathematical representation) to follow it. That's essentially it. That mathematical representation can then be relayed to the user.

Given the input I had above, using Claude Haiku 4.5, the math determines the most likely response is:

Truth and falsity do not apply

The LLM is not incorrect, or false, or wrong. The input asked for the most likely output and given that input, this is the most likely output. Truth and falsity has no bearing on the output of LLMs, the answer may correspond to reality or not but that is a matter of coincidence. The LLM itself is correct about the actual question, which is "Given this block of text, what is the most likely output?". That is the only question LLMs can answer, because that is what the math is about and as a math formula it never makes a mistake. You are not ever actually asking an LLM anything. When you send a message to an LLM you are adding a piece of text to that input (along with the system prompt generally determined by the provider) and are essentially saying "Given all this input, what's the most likely output" and the LLM gives you the most likely output. Your question is just used to build that input block, the LLM has no awareness of it as an actual question.

If you have a mathematical function called let's say blus that for odd numbers works just like plus, but for even numbers it adds an additional 1 such that:

Blus is not false, or incorrect. It is exact, the mistake is in thinking that blus is equivalent to plus. LLMs are essentially the same, they are mathematical formulas producing an output. That output may like blus appear to map onto reality often (if you only used blus on odd numbers, it would appear identical to plus) however blus is just a function that changes one number into another. An LLM is a function that transforms some text into other text. Correctness or truth is not a question, it merely transforms it and different transformations may be more or less useful for different situations.

Just as 5 is the exact answer of 2 blus 2, the fake description of weather in Portland is the exact answer the LLM provides. The output is not an answer to your question, the output is given the input as a whole what is the most likely text to follow from this. It is correct that given its dataset and that input that the most likely output is its description of fake weather in Portland. It is correct if you keep in mind what an LLM actually is, and the fact you do not actually ask anything a question. You merely provide input to which it responds with what it mathematically determines is the most likely output.

There is no consciousness, no intelligence, no wisdom. There is no AGI and no AGI will ever come from this. All LLMs do is given some input, generate the most likely output. There is a huge degree of marketing it to make it seem as though it is more than that, however we think it is important to be clear about what is actually going on.

How we use AI

Our usage of AI reflects this. We do not give AI any authority or the ability to determine anything concrete. Despite possibly seeming quite negative above, AI is still incredibly powerful and the ability to generate output based on any natural language input is extremely useful.

The way you will see AI used in our services is:

1. Generating possibilities and personalized options for the user (given the user's data + what we are looking for, what options does the LLM think are most suited probability wise?)
2. Providing engagement and reflection by responding to the user's action and input (you fill out a form and it shows up in a chat offering reflection about how your input relates to your previous history)
3. Being triggered at scheduled times to assess the user's data to see if there is some additional input the user could provide (you said you were going to do something, but it's been a week and you haven't done anything, ask the user about it)

AI is exceptional for each of these and doing these would just not have been practical before.

Why we make AI's nature apparent

A main priority throughout all of this, as is evident through this page, is to help our users to understand what AI actually is. There is no escaping it now and being mistaken about what it actually is can be harmful. Many areas where AI are used do the opposite of our approach. They obscure that an AI is present or try to make it seem human, and in order to avoid the negative effects of that we seek to make the nature of AI more apparent so people are less able to be mistaken about it. One of the features we have to do this is we allow the user to select from a variety of AI personalities and encourage them to customize it further on their own. The goal of that is to make it clear to the user that the output they are seeing is a product of something they helped construct.

AI is an incredibly useful tool, however understanding it and building the tooling to leverage it is quite complex. One of the main goals of Windy Hill Softworks is to make some of the benefits of AI more accessible to people, while also ensuring they can maintain an awareness of what AI is. We aim to give people access to the benefits of AI in a way that also gives them a better understanding of what AI actually is.

If you have any questions about our usage of AI or this policy you may contact us at contact@windyhillsoftworks.com.

Some concepts involved in how LLMs work were simplified to make this more accessible. This is not intended to be an exhaustive case, but more of a description of the approach we take to AI and our usage of it.