Can a neural network write code?

Question

Neural networks

I was interested in the topic of whether the neural network itself can write code. At first I didn’t find anything sensible in Google, then I came across this video . Can you tell me if the video is true? Is such a neuron already possible?

Answer 1

A girl comes to get a job.
- I work on a computer and type 1200 beats per minute.
All employees in amazement:
- What speed!
And then quietly to the side adds:
- But such nonsense is obtained!

I remember there was even a startup - supposedly the neural network generated the code.
But under the hood, it turned out that the Indians acted as a neural network.
The cost of equipment and time to create and train such a network is still higher than the cost of programmers to write code of at least comparable quality. ...

Answer 2

The question is rather philosophical and has more to do with programming languages ​​than with neural networks.
Let's start from the very beginning. How will the task statement for the neural network sound?
For example: write the string "hello world" 10 times with a pause of 1 second between them.
Okay, the neural network can handle this quite well. In Ruby, you get something like: But wait, this is almost the same as what we wrote in human language! Here the second question arises - what should be the level of detail of the task so that the level of detail of the code corresponds to it? And then, after a little thought, we understand that programming languages ​​are the same neural network - an interpreter or compiler as a bridge between the human description of the task and its machine implementation.
10.times { puts 'hello world'; sleep(1) }
We complicate the task. A person says to the neural network: I want a game with a ball that jumps around the screen, bounces off walls, breaks a wall of bricks, and the user controls the board and beats it off.
It is clear that a person immediately imagines the Arkanoid game in his mind. But the neural network has no such experience! This is where the details begin. How fast is the ball flying? Does the ball disappear if it is not hit by the board? Is there a second chance? And the third? What if all the bricks are broken?
All programming consists of thousands of small requirements questions that we don’t even think about when we write code. And how should the neural network learn about all this? And again, we come to the conclusion that having reached the maximum level of specification of the requirements, it turns out that we have written a program in natural language and it will not be difficult to translate it into machine language. And if so much work has already been done, why shouldn't a person himself code this task right away in a programming language?

Answer 3

Yes, it can .... Alice can already color the pony on command with her voice .
All you need to
Connect speech recognition ....
then drive in ready-made blocks. and then you can write a program with the help of a neural network with your voice
....
The only problem is that the mouse is really faster
. the module that will draw the field You say draw the field ... The neural network will take the block of the field and draw the field This is how it works. But you must have a ready-made template base from you.

The neural network will understand that we are talking about the field and the sides ....
And it will draw ..
To be honest, you can create about 120 templates according to which the game is built, this is enough to create any game
And then on request
I want a game with the Indians, step by step and a neural network based on 120 templates will create a game for you.
It's just that this game will be similar to the existing 120 templates.
It will create a program, but it will not differ much from everything that already exists.

Answer 4

no
, but if you throw it on a card or a crypt, I’ll tell you what, yes, and fill in what you need