I love programming games. Sadly, this is not a programming game. I struggle to even call this a puzzle game when so many of the levels are putting a single piece in the middle of the board and connecting its ins-and-outs to the board.

Buying better hardware means you'll run the same solution you had on an earlier level and upgrade from a silver star to a gold star - not because you made anything better, but just because "your computer is now faster".

Levels randomly restrict which nodes you can use so while they added the ability to reuse your solutions from previous levels, odds are that level used a node you're not allowed to use and you'll just have to do everything from scratch. Except for that one level where it forces you to use a previous level's solution.

Late game pieces are not fun to play with - because the game grades you on "processing time" you get faster pieces with small error chances. The game expects you to briefly hook these up to your system with an extra "gradient descent" node to "teach it", but once you do this, just copy that node to all the places you need it to avoid "retraining" them.

The worst offender is, sadly, the driving simulations levels. They claim to use a genetic algorithm to evolve the AI. The solution to all of them is to just click the evolve button once and everything's done. Each level adds more sliders which tune the way in which your AI evolves - but you just ignore them and click the evolve button. Later levels add side-facing LIDAR so your car's AI has more information about when you pass another vehicle! Except this only confuses the poor AI and removing the side LIDAR solves everything.

I played it to the end to try and see what it had to offer. I'm sad to say that past the first hour or two, you'll have seen everything of value in the game.

Pros:
- The graphics and sound are clean. The production quality feels high.
- The visual editor for solving puzzles is easy to use. It's also easy to use solutions from past puzzles as nodes in your current solution.
- The game gives useful links to those interested in learning more about machine learning.

Cons:
- After a short time, the puzzles become very uninteresting. Most of them fall under the "load balancing" category, which means that you spend most of your time figuring out how to allocate output symbols to output streams to match the volume dictated by the puzzle.
- The test run does not always terminate when the test is unsolvable, for example if all inputs have been sent to output streams, but the current accuracy and / or volume in the streams is not adequate. In these cases, if you want to see what went wrong, you will have to wait for the test to time out until you can view the stats for each output stream. This is annoying in combination with the prior point, when you are trying to debug which stream is getting too many nodes so that you can balance the output.
- Most of the machine learning concepts introduced in the game end up being irrelevant to the puzzles. In particular, the reinforcement learning puzzles and RNN puzzles don't let you use the ideas to come up with clever solutions. The RL puzzles give you a lot of knobs to play with, but the puzzle doesn't require anything other than sending some inputs to the output stream. The RNN puzzles give you the idea of RNN, but don't allow you to connect the nodes in any interesting or meaningful way: each node has memory, but you are forbidden from doing anything with it other than feeding it back into the same node that produced it.

Overall I don't recommend this game. The puzzles stop being interesting very early, and those that attempt to introduce more complicated machine learning topics avoid the hard task of introducing it in a way that the user can learn about the topic via the puzzles.

I spent seven hours getting to the end of this game, and I'm of mixed feelings about it.

At its core, While True Learn is a puzzle game: given bins of input objects (red/green/blue circles/squares/triangles), move them to the output bins that want them (like, one bin might want red circles and blue squares). You do this by building a flowchart made of conditional blocks: a "Decision Tree: Color" can send one color to the top output and another color to the bottom output (the third color is sent randomly to either output), a "SIFT" sends one shape to the top and the other two shapes to the bottom, a "MARK-1 Perceptron" can divide the colors across its three outputs but requires training (and misidentifies some objects), and so forth. The challenge is to send the inputs to the correct outputs using a minimum of blocks and within a time limit, so you may have to spend some time optimizing your approach.

At the same time, though this game is all about machine learning. "Learn how machine learning really works!" says the game's blurb, and the puzzle descriptions sound right up this alley. Predict a presidential election! Sniff out industrial espionage! Optimize roadway traffic! Write a chess bot! Recognize UFOs! The blocks you use all are named after real-world machine learning algorithms and have similar strengths/drawbacks. As you progress through the game and are introduced to new kinds of blocks, the game provides you with links to Wikipedia articles and YouTube videos explaining them and their history.

Problem is ... none of that actually has any bearing on the puzzles, which are still only about moving red/green/blue circles/squares/triangles from input bins to output bins. Whether you're identifying viruses or writing a movie rating system, you wouldn't know it from the puzzles themselves, which all look the same. Some of the puzzles (even in later levels) are trivially simple; others are too complicated to be fun. The only puzzles that really stand out are the ones about teaching a car to drive by writing a solution for it and adding LIDAR to the front and sides, but these still confused me - the car would accelerate until it rear-ended someone and then it would tailgate that car until the test was up, at which point I'd usually get a passing grade because the car's average speed was above the requirement.

On top of this are a strange array of features. You can buy hardware upgrades (after which simply re-running your earlier solutions can make them finish faster and get you more points), decorations for your office (the menu screen, which you don't spend much time on and you can only display a few decorations at a time anyway), and different kinds of cats (I didn't understand that at all). You can invest in startups, which means writing a solution for them and then clicking "Next Day" over and over again until the company stops making money. You can unlock "DLLs," which appear to be a way to re-use your previous solutions (I never used these because it was easier to write each solution from scratch), and "CatHub repos," which are just a notepad where you can try different approaches. There appears to be a well-developed "secret code" system where you can enter codes and unlock cosmetic items in the game.

All in all, it's really uneven. This could have been a much more fun puzzle game if it had focused on the puzzles, or it could have been a much more educational game about machine learning if it had made the puzzles resemble real-world problems. (Think of Kerbal Space Program.) It's enjoyable, but I don't think it stands out. This game teaches machine learning as well as Civilization teaches politics - which is not at all; you only get to use the end results. There are better puzzle games out there, and there are better simulator games.

The programming in this game is only present in its theme. You will not learn anything about machine-learning, programming, or algorithms. Watching the shapes fly past can be satisfying, but this game heavily boasts that it teaches machine-learning. If it didn't pretend it was more than a shape-sorting game, I probably wouldn't be leaving a negative review.

Sad to give this a negative review, but after 6 hours I just can't make myself continue.

1. It's not a programming game. And there's no machine learning other than names and links to Wikipedia articles. In fact, it's not even a puzzle game. It has things like node upgrades which means solutions are not deterministic: running them twice can yield different results. You can also unlock nodes in the shop to use in levels, which means solutions are even less deterministic and players can't compare.
2. I don't know how they did the RNG, but it's not right for a puzzle game. You can run the model, see where the inputs go, and then make a minor tweak only to find the distribution is totally skewed. What makes this worse is that the input sizes are absolutely tiny. So while a node may have a 50/50 of output, you may actually end up with 70/30. So any plan you had that works at statistically significant scale is ruined because it's 1 off. I've had 2 or 3 levels where my solution worked, but I had to move the ordering of nodes around to get the RNG to work right.
3. Side features. Training the "AI" (neither the manual training or the node training), startups,... they're not good. The mechanics aren't interesting or, worse yet, are poorly explained. What's fun about having to put down a node, hook it up to the input, then run "Train" and wait 20 seconds TWICE to get it to go down to 20%? The developers even seem aware of this because the sandbox area explicitly turns this feature off!
4. How the game presents it's levels is just... boring. In games like Shenzen or Infinifactory there's topology and each level has a unique characteristic to it. This game takes puzzles about filtering video game categories or movie ratings or pizza toppings and turns every single one into a completely unskinned "sort by 3 shapes and/or 3 colors".
5. There are quality of life issues. You can't adjust the input/output nodes on the board and they're unaffected by zoom - given the node limit, this doesn't constrain your solutions at all it just means you have to pack any complex solution together in a way that makes it unreadable. No orthographic lines or cross-through visuals like real diagrams use, so sometimes you can get wires crossed. Some nodes require an aforementioned annoying training process. The hitboxes for selection aren't constrained to the visual node, so you'll often select more boxes than you mean to for copy & paste or dragging. Puzzles can only speed up 3x, so you can waste 40 seconds testing solutions.

This game is not what it promises to be at all.

The puzzles are utterly braindead and they teach you nothing about machine learning concepts. You put a square peg into a square hole, you put a round peg into a round hole and voila! you now have software for a self-driving car!

Add to it the atrocious writing (I tell ya my duderinos, they couldn't make me cringe harder if they deliberately tried) and the single goofy song that repeats in a few seconds and you have a "puzzle game" that's hellbent on insulting your intelligence.

I guarantee you that the "senior software developers" who claim to have learnt something from this game in other reviews are LARPing. Because this game is a great power trip, you get to earn loads of money and get treated like a genius by doing absolutely nothing smart.

I may be less inclined to write a negative review if they marketed the game towards kids. But as it is, the description on the steam page is simply deceptive, there's no other way to put it.

This game showed potential, then got boring to finally be very disappointing and show major flaws.

First of all, forget all about the "simulation", "programming" and "education" tags tied to this game. Only the theme of the storyline has to do with machine learning and programming. The gameplay is just about simple puzzles, most of them solvable by trial and error.

Most puzzles in this game (there are exceptions, I'll talk about it below) are about one thing : sorting some data (as fast as possible). That data is represented by 3 possible shapes, each of which can come in 3 different colors. That's it. All the input will be 1 of 9 symbols.

At first, it's ok. Through the game you have to work with nodes to sort data, and those nodes do a bad enough job that it's challenging, you have to think. But once you figure how to sort the 3 colors, I'd say you already beat 75% of the game. The rest is way too simple. You keep repeating that sorting pattern to a point where it's boring.

So I did beat a few puzzles, then beat another one but couldn't get a gold medal for it. (Your grade is a bronze, silver or gold medal). I kept trying but no, that was beyond me. So I moved to the next puzzle. At one point I got tired of doing optional puzzles, so I moved through the story line as fast as I could.

The joke is, that gave me enough money to buy nodes upgrades, and I kept moving forward. Then beat the game. Finally, with all the upgrades I bought, I came back to the puzzles that gave me a tough time earlier. It turns out that in many cases, my old solution which was good enough for bronze/silver got me gold just because I now had enough upgrades.

That's bad. REALLY bad. You spend time trying to figure a creative solution, to later realise that you had it all right on first try but you needed more node upgrades.

All those sorting puzzles where the data is just 9 symbols are maybe 80% of the game.

Then there are a few challenges where you have to make an AI learn to drive a car. It pretty much comes down to clicking a "evolve" button about half a dozen times. That is not even a puzzle.

At the very end of the storyline, you have to sort letters instead of symbols. I thougt "Nice! Something different". Well, different yes, but nice, not at all. To sort the letters, you can only use one type of node. I daisy chained a few, had no idea how they worked, still did beat the challenge with gold medal. And the last chapter is all the same dumb thing, only you use more of the same node on later challenges. This literally is as dumb as a puzzle can be. And that's how the game ends.

There are also some "startup" challenges throughout the game, which could have been interesting. But they use the same mechanic as the regular sorting puzzles. Only bigger. And then I had no idea what was the optimal solution they were looking for. A fast one? A cheap one? One that throws no data to the trashcan? I had no idea. So I built a fast one, got trash revenues from it and that's it.

So I had fun for 60 minutes at the beginning. Then I got bored of trying puzzles over and over again where it was literally impossible to get a gold medal. So I rushed braindead through the game in a few hours. I was hoping for some different types of puzzle, but no. Then I came back to the early puzzles and got gold on all of them because I now had the required upgrades.

Bad game. Overpriced even if it would be sold for $2.

"while True: learn()" is not a programming game (such as Corewar, Tis-100, Human Resource Machine...).

It is only a eyecandy node and queue game (like as dump factory simulator for win98 or flash) with some sugar of historical data about machine learning. Remember you don't learn or play to coding a machine learning only do dumb puzzle things.

I paste the best metacritic review about this fake programming game:

Good artwork and interesting subject but "coding" in this puzzle game is as dumb as the manager of your company would expect it to be. Just link three functions and Boom! You made a disruptive startup... This interpretation of machine learning is more a marketing buzzword than something realistic.

It's.......... like something doesn't added up.

Before anything, this is NOT a programming game. Not in the conventional sense. It is, otherwise, a fuzzy-logic puzzle game. So prepare to be disappointed if you think you gonna learn some real-world programming concept from it.

It's also NOT a machine learning game either. Not in an understandable sense. Since everything you'll do is drag-and-drop built-in machine learning function (called node in this game) onto the screen and just wired them together to produce a desirable output. But the matter of HOW those functions work are left as an exercise to the player. Literally. Since at one point in the game, there's a message says:

Originally posted by while True: learn():

Learn how it works (Warning, Maths, integrals, derivatives and other scientific word inside)
LINK_TO_RANDOM_ARTICLE_OR_VIDEO_ON_THE_INTERNET

Math is already a hard subject, you don't have to make it scary (and isn't it YOUR job in the first place to try teaching these concepts via the game you create).

It's like you're start with a grand premise "learn how machine learning really works!", but then fail to even show what's behind the curtain.

For comparison, it's like saying "learn to build a car", then just give us a few simple LEGO System parts consisting of wheels, plates, and a minifigure!

Wheck! Some advance LEGO Technic sets even give us a lot of real-world-engineering details to build, such as V8 engine, differential, suspension, even sometimes a complex 8-speed gearbox!

That's fine. That's fine. Real world programmer blindingly implement machine learning with these ready-to-use tools in the same manner as this game anyway, right?

Well consider this. The game does measure things around speed, not accuracy.

Furthermore, the level design does revolve around the concept of three-star scoring. That is, build a machine that solve this level in 30 seconds to gain three stars, 40s for two, 50s then just one. But if the machine take longer than 1 minute then mission fail!

Oh boi, how can it be more wrong than this...

Apparently, player can buy upgrades to make their computer faster. So it's not depend on player's BRAIN anymore in order to solve the puzzle efficiently.

Secondly, who the well drawn the line between how many stars should we get. A God???

The star system is so broken since it suggest one and only one thing: the developer already know what's the best solution to tackle on a specific level. So they design the level to be efficiently solvable based on their envision. That's break immersion from the real-world puzzle (research, if you may) that we don't even know if we can keep pushing for a guaranteed improvement or not!

Take a look at Zachtronics, they themselves don't even know the most optimized way to solve their own designed puzzle. Their job is just only ensuring that there're at least one way to solve a level. Then the rest is up to the leader-board generated later by the community. So build your machine as slow as you like if you just want to progress, or spend lots of time optimize it so you feel proud of yourself being in the top percentile on the histogram. It's all up to you!

Zach's games, despite branding as "fake programming", actually teach us real world programming concepts! Especially when you aim to go faster by a magnitude. You have to actually go learning about real world optimization techniques. For example, a loop unrolling, like Duff's device.

Thus bring us to the most unforgivable point arose from the star system this game implemented. Which is it sometimes force us to do premature optimization, seeing that the level only need to shave another couple second to succeed.

What's programming aspects will players earn by this design? Not surprise if they're not learning anything. Instead they just fiddle around, rewire this to that, or even mess up with the PRNG's seed so that the puzzle might luckily be fast enough to satisfy the requirement. (Yet players will never know why it is faster.)

I'd rather say that's a dangerous seed (no pun intended!) to plant into the minds of us future programmer/people. Hence a deterministic thumb down.

A loved the idea but unfortunately this ends up being one of those games that tries to ride the line between educational and entertaining and fails at both. The game is less programming and more similar to the boring parts of being a network admin. You're just doing load balancing. A puzzle will want you to get 20% of the blocks into bin A, 30% into bin B, and 50% into bin C, and you have to arrange your splitters in such a way that the proportions come out right. When you get a "learning" node, it's just a splitter that makes errors, and to lower the error you have to run data through it. It doesn't need any particular data, or feedback, or to be arranged with other learning nodes in any way, you can literally just run random data through it and dump the data into nowhere and its error rate goes down. This is neither entertaining or educational.

There's also a mechanic for "upgrading your hardware" that really doesn't work well in the context of this kind of game. Grinding for better gear is okay in an RPG, but in a logical puzzle game where you're trying to find the exact optimal solution, you can be banging your head up against a wall trying to figure out why your solution isn't working and it turns out you got it right you just have to go upgrade your processor. Imagine playing portal and you're trying all kinds of angles and ideas to get up on some ledge, then it turns out you just have to go grind previous levels and buy 5 more jump upgrades so you can just jump up there, and that's the only way to beat it, and for the whole game there is no way to know if you're failing to think of something clever or you just need to muscle through it by buying more stuff.