Cybersecurity digital trends, mypy for code quality, Amazon's, Google's, Databricks', Alibaba' LLM announcements and Elon Musk creates mysterious company X

Apr 19, 2023

Welcome to my Effective AI newsletter, where I share my thoughts about AI, discuss latest digital news, as well as productivity and creativity tips for the AI community. If you missed my previous editions of this newsletter, you can check them out here. Additionally, you can visit my website for more of my articles, tips and tricks.

I also have this newsletter in the podcast form, so if you are interested in listening to it, please click on the player icon below.

🧠 What’s on my mind?

Cybersecurity digital trend

I recently posted my personal high level thoughts about top digital trends in 2023. In this article I would like to expand on the fourth trend: Cybersecurity. This is a short version and extended article can be read on my website.

In the year 2023, cybersecurity will remain one of the most important digital trends, especially in the wake of recent high-profile attacks on companies such as Optus, Medibank, Lattitude and Canva which have led to millions of personal data records, including driver license numbers and passport details, being stolen.

One of the key themes in cybersecurity for the coming year will be the automation of security processes through DevSecOps.

Another aspect of cybersecurity in 2023 will be the standardization of security best practices.

Trust architectures and digital identity will also be important trends in cybersecurity in 2023. The evolution of zero-trust architectures, digital identity solutions, and the development of protocols and controls for digital solutions will provide greater security for interconnected systems.

Lastly, there will be a further leverage of AI technologies to catch anomalies in systems.

🛠️ Practical tip - mypy for code quality checks

One of the best ways to enforce code quality in Python projects is to run mypy checks at commit stage and automatically in CI/CD pipeline. Let me explain why I recommend mypy.

Python is a dynamic programming language, meaning that you do not specify types for your variables but values in memory do have a type. So variable x can be both a number 123 and a string ‘abcd’ even within the same function. While it can be convenient not to specify types and reuse the same variable for multiple types, some patterns may hide serious errors.

Python 3 introduced type annotations, where developers can optionally specify types. For example

def my_function(x: int) → int:
     return x+1

In this case it is clear that my_funcion accepts int and return an int. The compiler ignores this type annotation. However, tools like mypy take it into account and tracks if we ever pass non-int type to my_function.

Mypy is an optional static type checker for Python and it analyses the code, similar to what compiler language would do. Initially it started as a standalone Python variant but over time it has been rewritten as Python 3 checker instead.

I usually recommend to start using mypy or similar checker as early in the project as possible, so that the team writes quality code and gets used to mypy from day 1, or as early as practical.

When you use mypy on an existing project it can be quite painful to fix all the errors and put type annotations in all files. However, this exercise almost always fishes out a few very hard to find bugs.

🌏 What’s happening in the world of AI?

Microsoft DeepSpeed

Microsoft released DeepSpeed, a framework that helps to optimise deep learning training with 15x speedup over state-of-the-art RLHF systems.

Databricks' Dolly 2.0

Databricks' Dolly 2.0 is now fully open-source and available for research and commercial use, boasting Instructed-Tuned LLM capabilities. It is instruction tuned LLM, with 12B parameters, trained for a fraction of ChatGPT cost.

The model is available on Hugging Face and the 15k dataset here.

Amazon joins the GenAI race

This week Amazon joined the GenAI race by announcing their generative models, Bedrock and Titan, CodeWhisperer - AI development co-pilot and new EC2 instances.

Alibaba is launching its generative AI model but CCP delayed the public access

Alibaba rolled out its Generative AI model, called Tongyi Qianwen. It can summarise chats, transform charts into apps and is integrated in Alibaba’s voice assistance. However, a few hours after announcement, China’s leading internet authority asked to register the model with them and provide additional information. This delayed the launch.

Anthropic raises capital to compete with OpenAI

Anthropic is another potential OpenAI contender, which is aiming to raise up to $5 billion USD. It is started by former OpenAI researchers. They want to build 10 times more powerful AI that the existing ones. Anthropic describes their model as a “next-gen algorithm for AI self-teaching” and refer to constitutional AI principles.

Elon Musk is to launch OpenAI rival - X

Elon Musk is to launch OpenAI rival, called X.ai. Most of details are unknown other than it distributed about 100 million shares and launched in Nevada.

Google is launching Project Magi to compete with Bing

Google is working on Project Magi, that will deliver ChatGPT like experience for Google search. Samsung considers switching to Bing and it will be big hit to Google, so Magi is very high on priority list.