Prototype: github.com/cirosantilli/Urho3D-cheat

Prior art research: github.com/cirosantilli/awesome-reinforcement-learning-games

Less good discrete prototype: github.com/cirosantilli/rl-game-2d-grid YouTube demo: Video 1. "Top Down 2D Continuous Game with Urho3D C++ SDL and Box2D for Reinforcement learning by Ciro Santilli (2018)".

The goal of this project is to reach artificial general intelligence.

A few initiatives have created reasonable sets of robotics-like games for the purposes of AI development, most notably: OpenAI and DeepMind.

However, all projects so far have only created sets of unrelated games, or worse: focused on closed games designed for humans!

What is really needed is to create a single cohesive game world, designed specifically for this purpose, and with a very large number of game mechanics.

Notably, by "game mechanic" is meant "a magic aspect of the game world, which cannot be explained by object's location and inertia alone" in order to test the the missing link between continuous and discrete AI.

Much in the spirit of gvgai, we have to do the following loop:

The question then becomes: do we have enough computational power to simulation a game worlds that is analogous enough to the real world, so that our AI algorithms will also apply to the real world?

To reduce computation requirements, it is better to focus on a 2D world at first. Such world with the right mechanics can break any AI, while still being faster to simulate than a 3D world.

The initial prototype uses the Urho3D open source game engine, and that is a reasonable project, but a raw Simple DirectMedia Layer + Box2D + OpenGL solution from scratch would be faster to develop for this use case, since Urho3D has a lot of human-gaming features that are not needed, and because 2019 Urho3D lead developers disagree with the China censored keyword attack.

Simulations such as these can be viewed as a form of synthetic data generation procedure, where the goal is to use computer worlds to reduce the costs of experiments and to improve reproducibility.

Ciro has always had a feeling that AI research in the 2020's is too unambitious. How many teams are actually aiming for AGI? When he then read Superintelligence by Nick Bostrom (2014) it said the same. AGI research has become a taboo in the early 21st century.

Related projects:

Related ideas:

Bibliograpy:

If you are a pussy and work a soul crushing job, this is one way to lie to yourself that your life is still worth living: do one cool thing every day.

Find a time in which your mind hasn't yet been destroyed by useless work, usually in the morning before work, and do one thing you actually like in life.

Work a little less well for you boss, and a little better for yourself. Ross Ulbricht:Selling drugs online is not advisable however.

Even better, try to reach an official agreement with your employer to work 20% less than the standard work week. For example, you could work one day less every week, and do whatever you want on that day. It is not possible to push your passion to weekends, because your brain is too tired. "You keep all non-company-related IP you develop on that time" is a key clause obviously.

On a related note, good employers must allow employees to do whichever the fuck "crazy projects", "needed refactorings or other efficiency gains" and "learn things deeply" at least 20% of their time if employees want that: en.wikipedia.org/wiki/20%25_Project. Employees must choose if they want to do it one day a week or two hours per day. One day per month initiatives are bullshit. Another related name: genius hour.

Highly relevant on this topic: Video "What Predicts Academic Ability? by Jordan B Peterson (2017)".

Maybe you will be fired, but long term, having tried, or even succeeded your dream, or a one of its side effects, will be infinitely more satisfying.

The same goes for school, and maybe even more so because your parents can still support you there. Some Gods who actually followed this advice and didn't end up living under a bridge:

Gandhi TODO source:

Ciro Santilli's full birth name is "Ciro Duran Santilli", with mother's last name "Duran" in the middle as per Brazilian tradition.

But Ciro's usage of "Duran" got gradually dropped to "Ciro Santilli", Ciro's official Italian name, as Ciro moved more and more definitively to Europe.

It can still however be seen in certain online places where Ciro didn't have the patience or power to change it e.g. some old École Polytechnique stuff: gitlab.binets.fr/ciro.duran-santilli/china-dictatorship

Also, don't have multiple names if you can avoid it, it is confusing!

See: Section "Ciro Santilli".

Stack Exchange solves to a good extent the use cases:points of view. It is a big open question if we can actually substantially improve it.

Major shortcoming are mentioned at idiotic Stack Overflow policies:

Bibliography:

This is mostly stuff from before 2016 when Ciro was anxious to document his contributions to get a job.

Most of the projects here are also minor contributions, or Ciro later noticed that the projects were not useful enough to work on and that he was actually wasting his time.

Not end-to-end encrypted by default, WTF... you have to create "secret chats" for that:

You can't sync secret chats across devices, Signal handles that perfectly by sending E2EE messages across devices:This is a deal breaker because Ciro needs to type with his keyboard.

Desktop does not have secret chats: www.reddit.com/r/Telegram/comments/9beku1/telegram_desktop_secret_chat/ This is likey because it does not store chats locally, it just loads from server every time as of 2019: www.reddit.com/r/Telegram/comments/baqs63/where_are_chats_stored_on_telegram_desktop/ just like the web version. So it cannot have a private key.

Allows you to register a public username and not have to share phone number with contacts: telegram.org/blog/usernames-and-secret-chats-v2.

Has Reproducible builds on Android and iOS: core.telegram.org/reproducible-builds

Self deleting messages added to secret chats in Q1 2021: telegram.org/blog/autodelete-inv2

Can delete messages from the device of the person you sent it to, no matter how old.

TODO why it is optimal: physics.stackexchange.com/questions/149214/why-is-the-carnot-engine-the-most-efficient

In 2020 Brazil for example, you are not allowed in theory to obtain a double nationality which you were not allowed to have as a birth right.

This means that Brazilian students e.g. in France, many of whom could easily obtain the French citizenship had to either chose between:Can you guess which option Brazilian students would usually pick?

As a poor country, you have to allow people to obtain multiple citizenship. Students are not going to go back because they don't have the foreign citizenship. They are just going to have to ensure shittier jobs for a few years, thus diminishing the chances that they will actually lean anything useful to bring back to your country later on.

You have to bring students back after they're masters with the carrot, and not with the stick.

Let's start with the one dimensional case. Let the $q:\mathbb{R}\to \mathbb{R}$ and a Functional $I$ defined by a function of three variables $F:{\mathbb{R}}^3\to \mathbb{R}$:

Then, the Euler-Lagrange equation gives the maxima and minima of the that type of functional. Note that this type of functional is just one very specific type of functional amongst all possible functionals that one might come up with. However, it turns out to be enough to do most of physics, so we are happy with with it.

Given $F$, the Euler-Lagrange equations are a system of ordinary differential equations constructed from that $F$ such that the solutions to that system are the maxima/minima.

In the one dimensional case, the system has a single ordinary differential equation:

By $\frac{\partial F}{\partial q}$ and $\frac{\partial F}{\partial \dot{q}}$ we simply mean "the partial derivative of $F$ with respect to its second and third arguments". The notation is a bit confusing at first, but that's all it means.

Therefore, that expression ends up being at most a second order ordinary differential equation where $q$ is the unknown, since:

Now let's think about the multi-dimensional case. Instead of having $q:\mathbb{R}\to \mathbb{R}$, we now have $q:\mathbb{R}\to {\mathbb{R}}^n$. Think about the Lagrangian mechanics motivation of a double pendulum where for a given time we have two angles.

Let's do the 2-dimensional case then. In that case, $F$ is going to be a function of 5 variables $F:{\mathbb{R}}^5\to \mathbb{R}$ rather than 3 as in the one dimensional case, and the functional looks like:

This time, the Euler-Lagrange equations are going to be a system of two ordinary differential equations on two unknown functions $q_1$ and $q_2$ of order up to 2 in both variables:At this point, notation is getting a bit clunky, so people will often condense the $q_i$ vectoror just omit the arguments of $F$ entirely:

These are the final equations that you derive from the Lagrangian via the Euler-Lagrange equation which specify how the system evolves with time.

The key difference from Lagrangian mechanics is that the Hamiltonian approach groups variables into pairs of coordinates called the phase space coordinates:This leads to having two times more unknown functions than in the Lagrangian. However, it also leads to a system of partial differential equations with only first order derivatives, which is nicer. Notably, it can be more clearly seen in phase space.

Bibliography: