An encounter with anti-gravity.

I thought to document this, i suppose as evidence.
First thing to do is read the terms contract. It's first and only line of prohibition is "using AI-generated content from our services to develop machine learning models or related AI technology." Well, panic sets in. So i asked Google[.]com/search it refused to budge beyond the few words of the contract. So i pleaded with the "ai mode" of google.
it said:

What about using Google Gemini to create code as a client for Ollama.com? Using AI on Google Search to create code for a client of Ollama.com is generally allowed, provided the code is for application development and not model development. Allowed and Not Allowed Uses Allowed: Application Development. AI on Google Search can generate code for software applications. This includes a client that works with Ollama's API or manages local models. Google does not claim ownership of the generated code. Not Allowed: Model Development. The terms prevent using AI-generated output, including code, to train, refine, or evaluate machine learning models that compete with Google's services. This means AI on Google Search cannot write the internal training algorithms for a new model or create synthetic training data to improve a model hosted on Ollama.

I persist, being banned from facebook by some (we wont say) creatures, has me a little worried about these fucdaic rules, that everyone would naturally assume is the only thing to do. You don't sow wheat seeds to grow watermelon icecream. Trade came from the spice route, i.e: people traded seeds. They didn't go around conquering the world to control all pepper in the world because it was easier with their ships and fancy leather bags. (as though AI is supposed to be used for non-AI)... so i asked again.

But isn't "Machine learning models" ubiquitous? couldn't it include everything including hand coded bayesian networks? While "machine learning" is a broad term, Google's Terms of Service are typically interpreted through a commercial and competitive lens rather than a strictly mathematical one. Is it Ubiquitous? Yes, technically many algorithms fall under the umbrella of "machine learning," including hand-coded systems like Bayesian networks when they are used to "learn" from data. However, the core of Google's restriction is distillation—preventing you from using their high-end models to train a cheaper or rival system. Hand-Coded Bayesian Networks A Bayesian network can be classified in two ways, which leads to different outcomes regarding compliance: Static Inference (Generally Allowed): Using the Gemini family of models to write the structure of a Bayesian network (e.g., "Write Python code for a Bayes net representing medical symptoms") is considered application development. This is generally allowed as it is a specific software implementation. Automated Learning (Restricted): Using the Gemini family of models to generate massive amounts of synthetic data to parameterize or "train" that Bayesian network so it can make predictions without Google enters the territory of developing a machine learning model. The Practical Rule of Thumb The terms distinguish between using the AI to build software and using the AI's intelligence to replace its own underlying model. Safe: "Write a script that uses an existing ML library to solve X." Violative: "Generate 10,000 examples of X so I can train my own model to do what you do". For most developers creating clients or logic-based tools, this restriction is rarely an issue unless the main goal is to clone or compete with the intelligence of the model itself.

Still seems silly given that there are quotas in place. just let the people use their quotas!! basic universal income, etc.

You WANT TO REVIEW everything. Antigravity is not ASGI... 

When you install Antigravity, set everything to you have control. Then you can only blame yourself when the statistical machine bumps around, like a blind surgeon within your preexisting code. 

Google has provided enough warnings, just accept the warnings as 'dont let the IDE ruin your code or your file system!!!.' 

GRAVITY, Sweet and Sour Gravity. 

A person needs gravity to excercise, to become proficient and better than the machines, A person needs their own source of gravitational potential energy. 

<iframe frameborder="0" height="270" src="https://youtube.com/embed/A8_nNYLTXEQ?si=jOglZOIkIAaNUFdV" width="480"></iframe>

BACKUP!!! 

Save your files. Get a new computer dedicated for this. Save everything. before you install. 

How to Do a project: 

1) Create a folder. 

When you create a project, your antigravity IDE (integrated development environment) will first search this folder for the relevant files. If you do not have your file in this folder, then antigravity will search your file system !! and once it has found the file, even if you later copy the file into the antigravity's folder, antigravity will still edit your file system's file. 

Lesson:  Before you start talking about a project, i.e: with the "Run and Debug File" button. Place the file inside your `antigravity` folder. 

Well now i've built a website, used AntiGravity to port my little HTML apps into PHP. So.. the following are some necessary things to know. If AntiGravity seems to stall, i simply assure it that my project is not for a competing AI. 

2) Place your project files in the folder. 

Let's say you created a program, and you want some help from AntiGravity. Then pop those files into the folder. Simple. It is better to keep a folder for a new project. If you have multiple projects within the same greater scheme. 

3) First explore the project

You can ask Antigravity to explore and understand the suite of files, and produce a document that fully explains and summarizes the nuances. If you have any nuances in mind, the structures, algorithms, details, ask AI to explore and detail those features of your project as well. 

4) Creative prompts should use the Chain of Thought format, with the 7 stage format for prompts as well. 

I like to use keywords to demarcate the sections of the prompt. 
Goal:
Subtask 1: 
Subtask 2: 
Abstraction: 
Many-Shot-attempt: 

The many-shot attempt is helpful if you want to include a feature from a different project into this project. It serves as a sort of template or example. 

5) Referencing the Editor Window. 

You can highlight a section of the code in the editor window, and right-click (or Alt + L) which creates a line reference into your prompt. 

So you can highlight a point of fault, say through using GDB,  and then ask AI to recognize that the problem is this line, and suggest reasons why the fault is occurring. 

You can reference multiple lines in the same prompt. All cushioned between your own text explanations. This gives you great ability to describe the code ,  like a control flow chain, describing hops and synonyms accross different scopes and files. 

 6) Proceed

To get Anti Gravity to work autonomously, you have to agree with its planned course of action. This plan is in the implementation.md file and the plan.md file. 

You will find little speach bubbles alongside each line or section which allows you to edit the implementation and plan. This means that you should take the due diligence to read and comprehend its plan, and take the time to think outside of the box, to see what it can't see. Sometimes the AI simply will do what you say, which is great, but it wont do the things that you assume will happen. 

This might require a great deal of thinking on your part, planning, fleshing out the 'philosophy, architecture and engineering' of your project.  I have created a number of tools to help me try to flesh out a project that is purely abstract, before it is created. 

But, if you are impatient, and the AntiGravity is not displaying the submit button, look for the PROCEED button in the implementation.md window. 

7) Walkthrough.md

Check the walkthrough.md when you get one, as AntiGravity will give you a little set of tests to see if its implementation is as you asked. Note: that it is here that you will realize that you didn't say what you assumed would be known by someone listening to you. 

8) Git hub Commits (Concurrent Versioning System)

I find github complicated compared to the old CVS of OpenBSD... Perhaps I am stupider now than back then. So I simply save a file with a prefix, when i create a change (like File_001HasMemory.h , File_002HasObjects.h ) 

In anycase, It is wise to save your progress before ruinning your hard work, heretofore.