True Positive Weekly #1

The most important artificial intelligence and machine learning links of the week.

Hey!

You receive this newsletter because you subscribed to receive AI and machine learning updates on the companion websites of Andriy Burkov’s two books: The Hundred-Page Machine Learning Book and Machine Learning Engineering.

There are about 15,000 subscribers now, so we, True Positive Inc., the company behind Andriy Burkov’s publishing initiatives, decided to start sending weekly updates. Let’s see how it goes. We hope you will enjoy it. Please don’t hesitate to send us your feedback at hello@truepositive.ca.

In this issue: a powerful antibiotic was discovered using AI, the GPT-2 model explained step by step, a guide for optimizing machine learning infrastructure cost, an overview of deep learning for text, and more.

• AI doesn't always take jobs, it just makes them suck

• A powerful antibiotic has been discovered using artificial intelligence

• A step by step explanation of the GPT-2, one of the most powerful pre-trained models for natural language processing

• A guide for optimizing machine learning infrastructure cost on AWS

• How art can help us understand AI

• Machines still can't work out cause and effect

• A step by step explanation of the GPT-2, one of the most powerful pre-trained models for natural language processing

• Deep learning for text: an overview

• Language models are few-shot learners (paper, discussion)

• A collection of 545 datasets for natural language processing

• TextAttack: a Python framework for model training, adversarial attacks, and data augmentation in natural language processing

• Paper explained: Discovering Symbolic Models from Deep Learning with Inductive Biases (video, paper, code)

---

Enjoy the newsletter? Help us make it bigger and better by sharing with your colleagues and friends. If you received this issue from a friend, you can subscribe to the newsletter here.

Have a nice weekend! See you next week.