Welcome back to another episode of Tech Bytes.

Let me introduce you to Mojo, a brand new programming language for all AI developers.

Python is the current most widely used language for AI developers, but Mojo is here to take the crown.

It is said to perform up to 35, 000 times faster than Python.

And you can also use the cute fire emoji as the file extension if that appeals to you.

If Mojo is actually 35, 000 times faster than Python, will Python be replaced by Mojo?

Before we get into how Mojo works, let's take a look at some characteristics of Python to understand how Mojo is different from it.

Python is currently the most widely used programming language for a lot of web developers, data scientists, data analysts, and more.

It's also the most famous choice as the first programming language to learn for beginners.

Python is considered a high-level language because it is fairly similar to the English language and is thus quite easy for humans to write and understand.

Another reason why Python is currently considered the best language for AI and machine learning is because of its huge libraries and frameworks.

NumPy, Pandas, TensorFlow, Keras, and PyTorch are some of the most widely used Python libraries for data science.

If Python is easy to learn and has all the libraries and documentation for machine learning, why did Mojo developers think that we need a replacement.

Let's take a look at some of the downsides of using Python.

With the same algorithm, Python is usually much slower than C, which is the language Python is written in.

Python is a dynamically typed language, which means Python variable types are not specified, and the codes are interpreted and compiled during their runtime.

This makes the code both easier to write but also slower to run.

Due to the slow running time, although most AI research is done in Python, most products end up being programmed in C or C++.

Another major downside to using Python is that it is difficult to distribute.

When a C programmer wants to distribute a C program, they can simply compile the program into a single file for distribution.

Now, let's say a Python programmer wants to distribute a Python program using the Keras library.

The Users would have to download Python, Keras library, and all the other dependent libraries to run the file.

This is also mainly because Python has to be interpreted and compiled during this runtime.

This is where Mojo comes in, with features that resemble Python but could tackle the downsides of it.

Mojo is created by Chris Ladner, a legendary engineer who has led several projects at Apple, Tesla, Google, and more.

Mojo is not open source yet, but will soon be after its official release.

To try out Mojo, you can sign up to the waitlist.

You will be able to get access to Mojo in only a few hours after being on the waitlist.

Now let's take a look at how Mojo is currently different from Python and how the developers managed to make it 35, 000 times faster.

Mojo is a superset of Python, just like how TypeScript is a superset of JavaScript.

This means that Mojo uses all the Python syntaxes and libraries while bringing the speed up to 35, 000 times faster.

Mojo is fully compatible with the Python ecosystem.

You can use all the existing Python libraries that were previously mentioned and so much more.

You can easily import Python libraries with the syntax python.importmojo.

Another very fun feature of Mojo is that it supports the fire emoji to be the file extension.

You can either name the same file hello.mojo or hello.fireemoji.

Now we know that Mojo supports some fun emoji extensions and that its syntax is very similar to Python.

Then let's take a look at some of the features of Mojo that makes it faster.

To be precise, Mojo being up to 35, 000 times faster than Python doesn't mean that the same Python code will magically run 35, 000 times faster in Mojo.

With the exact same code, Mojo says it runs only slightly faster than Python.

These are some of the differences that you have to make in your code for it to run significantly faster.

This is an example of a function declaration in Mojo.

It looks exactly like a Python function definition except for a few differences.

Instead of def, Mojo function declarations use fn.

Also, the types of parameters and the return variables have to be specified.

This switch the code from dynamically typed to statically typed.

Mojo's objects can be both built in a class or a struct.

A class is also supported in Python, but a struct is added in Mojo.

Similar to a class, a struct can have variables and functions defined in it.

The difference again is that variables are static rather than dynamic.

In this piece of code, you can see that the 2 variables first and second are set as integers.

It also contains the initializer and a user-defined function that prints out the variables.

The user-defined function dumps the fn method that was introduced previously.

Moreover, Mojo has 2 types of variable var and let.

Let's take a look at a simple code snippet as an example.

The simple main function increments the variable x after setting it.

If you change the var to let, however, you will be getting an error like this.

This is because let makes the variable X immutable, so an attempt to increment it would lead to an error.

Using static types like this can optimize running time and help in error checking.

But you can still use the dynamic types like Python if it is not necessary.

Mojo also allows parallel processing with the use of a multi-level intermediate representation compiler, MLIR in short.

Python only allows single-threaded executions, so many AI programs had to use CUDA as a roundabout way for parallel processing to shorten the runtime.

However, with MLIR, it's now possible to scale to exotic hardware types without too much complexity.

This could become extremely handy since many machine learning programs are currently using CUDA to reduce the often extensive training time.

Mojo adopts the ownership and borrowing concept from Rust.

The ownership and borrowing concept is a unique memory management feature in Rust.

Python has a garbage collection feature that regularly looks for no longer used memory as the program runs.

In C and C++, the programmer must explicitly allocate and free the memory.

Ownership and borrowing is the third approach that could be a lot more convenient for programmers who are familiar with the concept of stacks and heaps and will not slow down the program while it's running.

Another memory management feature that Mojo has is the pointers from C and C++.

This could be used for more low-level customization.

Pointers could be employed with the command import pointer.

Finally, the deployment of a program file is much easier in Mojo than in Python.

I've explained previously how Python codes are interpreted and compiled in runtime.

On the other hand, Mojo is compiled ahead of time to machine-native code.

In other words, program files are packaged nicely into a single file and the users wouldn't have to download Python and all the dependent libraries for them to run.

We've discussed the new features of Mojo that were added on top of its Python-compatible syntaxes and how these new features could dramatically shorten the running time and make the deployment easier.

But will it actually replace Python or will it just be another programming language that's never actually been used and be forgotten?

Although Mojo is much faster than Python, there's still a trade-off between runtime speed and code complexity.

This means that while Mojo could be faster, the code will be much more complex than normal Python codes.

For example, changing all the user-defined functions and variables from dynamic to static could improve the runtime but would also make the code significantly more complex.

While some say even if it's more complex, it could be easier for Python developers to learn compared to learning a whole new language like C++ or Rust.

This could perhaps fill in the gap between research, which is mainly written in Python, and development, which is mainly written in C or C++.

A lot of people are skeptical of how big Mojo could become, but programmers are still very excited with the news.

If you'd like to hear more follow-up stories on Mojo, or want to keep up with how machine learning is evolving day by day, Stay tuned for the coming episodes of TechBytes.

This was your host Ericsson, and I'll see you


Speaker 0: Welcome back to another episode of Tech Bytes. This is your host, Ericsson. Let me introduce you to Mojo, a brand new programming language for all AI developers. Python is the current most widely used language for AI developers, but Mojo is here to take the crown. It is said to perform up to 35, 000 times faster than Python.

And you can also use the cute fire emoji as the file extension if that appeals to you. If Mojo is actually 35, 000 times faster than Python, will Python be replaced by Mojo? Before we get into how Mojo works, let's take a look at some characteristics of Python to understand how Mojo is different from it. Python is currently the most widely used programming language for a lot of web developers, data scientists, data analysts, and more. It's also the most famous choice as the first programming language to learn for beginners.

Python is considered a high-level language because it is fairly similar to the English language and is thus quite easy for humans to write and understand. Another reason why Python is currently considered the best language for AI and machine learning is because of its huge libraries and frameworks. NumPy, Pandas, TensorFlow, Keras, and PyTorch are some of the most widely used Python libraries for data science. If Python is easy to learn and has all the libraries and documentation for machine learning, why did Mojo developers think that we need a replacement. Let's take a look at some of the downsides of using Python.

Most importantly, Python is slow. With the same algorithm, Python is usually much slower than C, which is the language Python is written in. Python is a dynamically typed language, which means Python variable types are not specified, and the codes are interpreted and compiled during their runtime. This makes the code both easier to write but also slower to run. Due to the slow running time, although most AI research is done in Python, most products end up being programmed in C or C++.

Another major downside to using Python is that it is difficult to distribute. When a C programmer wants to distribute a C program, they can simply compile the program into a single file for distribution. Now, let's say a Python programmer wants to distribute a Python program using the Keras library. The Users would have to download Python, Keras library, and all the other dependent libraries to run the file. This is also mainly because Python has to be interpreted and compiled during this runtime.

This is where Mojo comes in, with features that resemble Python but could tackle the downsides of it. Mojo is created by Chris Ladner, a legendary engineer who has led several projects at Apple, Tesla, Google, and more. Mojo is not open source yet, but will soon be after its official release. To try out Mojo, you can sign up to the waitlist. You will be able to get access to Mojo in only a few hours after being on the waitlist.

Now let's take a look at how Mojo is currently different from Python and how the developers managed to make it 35, 000 times faster. Mojo is a superset of Python, just like how TypeScript is a superset of JavaScript. This means that Mojo uses all the Python syntaxes and libraries while bringing the speed up to 35, 000 times faster. Mojo is fully compatible with the Python ecosystem. You can use all the existing Python libraries that were previously mentioned and so much more.

You can easily import Python libraries with the syntax python.importmojo. Another very fun feature of Mojo is that it supports the fire emoji to be the file extension. You can either name the same file hello.mojo or hello.fireemoji. Now we know that Mojo supports some fun emoji extensions and that its syntax is very similar to Python. Then let's take a look at some of the features of Mojo that makes it faster.

To be precise, Mojo being up to 35, 000 times faster than Python doesn't mean that the same Python code will magically run 35, 000 times faster in Mojo. With the exact same code, Mojo says it runs only slightly faster than Python. These are some of the differences that you have to make in your code for it to run significantly faster. This is an example of a function declaration in Mojo. It looks exactly like a Python function definition except for a few differences.

Instead of def, Mojo function declarations use fn. Also, the types of parameters and the return variables have to be specified. This switch the code from dynamically typed to statically typed. Mojo's objects can be both built in a class or a struct. A class is also supported in Python, but a struct is added in Mojo.

Similar to a class, a struct can have variables and functions defined in it. The difference again is that variables are static rather than dynamic. In this piece of code, you can see that the 2 variables first and second are set as integers. It also contains the initializer and a user-defined function that prints out the variables. The user-defined function dumps the fn method that was introduced previously.

Moreover, Mojo has 2 types of variable var and let. Var is mutable and let is immutable. What does this mean? Let's take a look at a simple code snippet as an example. The simple main function increments the variable x after setting it.

If you change the var to let, however, you will be getting an error like this. This is because let makes the variable X immutable, so an attempt to increment it would lead to an error. Using static types like this can optimize running time and help in error checking. But you can still use the dynamic types like Python if it is not necessary. Mojo also allows parallel processing with the use of a multi-level intermediate representation compiler, MLIR in short.

Python only allows single-threaded executions, so many AI programs had to use CUDA as a roundabout way for parallel processing to shorten the runtime. However, with MLIR, it's now possible to scale to exotic hardware types without too much complexity. This could become extremely handy since many machine learning programs are currently using CUDA to reduce the often extensive training time. Mojo adopts the ownership and borrowing concept from Rust. The ownership and borrowing concept is a unique memory management feature in Rust.

Python has a garbage collection feature that regularly looks for no longer used memory as the program runs. In C and C++, the programmer must explicitly allocate and free the memory. Ownership and borrowing is the third approach that could be a lot more convenient for programmers who are familiar with the concept of stacks and heaps and will not slow down the program while it's running. Another memory management feature that Mojo has is the pointers from C and C++. This could be used for more low-level customization.

Pointers could be employed with the command import pointer. Finally, the deployment of a program file is much easier in Mojo than in Python. I've explained previously how Python codes are interpreted and compiled in runtime. On the other hand, Mojo is compiled ahead of time to machine-native code. In other words, program files are packaged nicely into a single file and the users wouldn't have to download Python and all the dependent libraries for them to run.

We've discussed the new features of Mojo that were added on top of its Python-compatible syntaxes and how these new features could dramatically shorten the running time and make the deployment easier. But will it actually replace Python or will it just be another programming language that's never actually been used and be forgotten? Although Mojo is much faster than Python, there's still a trade-off between runtime speed and code complexity. This means that while Mojo could be faster, the code will be much more complex than normal Python codes. For example, changing all the user-defined functions and variables from dynamic to static could improve the runtime but would also make the code significantly more complex.

While some say even if it's more complex, it could be easier for Python developers to learn compared to learning a whole new language like C++ or Rust. This could perhaps fill in the gap between research, which is mainly written in Python, and development, which is mainly written in C or C++. A lot of people are skeptical of how big Mojo could become, but programmers are still very excited with the news. If you'd like to hear more follow-up stories on Mojo, or want to keep up with how machine learning is evolving day by day, Stay tuned for the coming episodes of TechBytes. This was your host Ericsson, and I'll see you

See all Preface transcripts on Youtube

New language Mojo.🔥 | Will Python be replaced?