A first look at OpenAI GPT2

GPT2 is a model developed by OpenAI for NLP.
The model is composed by a transformer architecture as shown in the picture below.

Luckily for us the dudes at Huggingface have releases a pretrained version and implementation of this model ... and many other type of transformer models.

https://huggingface.co/transformers/pretrained_models.html

I've created a small google colab notebook where you can try the large version of the pre-trained model to generate some text.

https://colab.research.google.com/drive/1Jlp_39bxUuWtaJfEFRZOK-5RshDPKlxw

The followings are some examples of generated text, the text in cursive is the text that has been used as input

AI talking about Docker

Docker is a type of virtualization service that enables users to create small virtual machines within the confinement of their OS. The term used to describe this type of emulation is often referred to as Component Based Emulation (CBE).

This article will describe the principles of Docker and explain the core concepts.
A single container should have a single process.

In order to achieve this you should have a single process running in the container, without any user input. If you have more than one process running in the container, you should manage the processes individually using the docker user management tool, with the built-in support for single containers.
The user of a given process should be as

AI talking about AI

Artificial general intelligence (AGI) is the intelligence of a machine that has the capacity to understand or learn any intellectual task that a human being can.

As we move forward, however, we will be required to begin to compose new kinds of AI using ever larger sets of code, thereby opening up many new scientific and social challenges. While our understanding of the fundamental issues that will determine the emergence of "advanced" AGI is too broad to form a thorough (e.g., evolutionary) picture, there are a few questions that deserve attention now. The first is the requirement that, for every new issue, we need a general framework for thinking about the possibility of there being a solution.

I.e., how do we choose the problem to be solved, what is the nature of the problem, and what qualities are necessary for this solution to be successful? We would also like to know what approaches would be appropriate if we were faced with the same problem. Second, we need to decide how to model the intelligence involved. The fundamental issue here is what type of AI will be necessary for AI to become an effective part of our lives.

From what I have read, we may be able to have a more general idea about this (and when we may be able to have a definitive answer). Third, we will need to decide what abilities we will need to have as an AI.

AI completing the answer to life the universe and everything

This is the answer to life, the universe and everything
, +10100000, where each letter represents one year since 1900. All numbers up to 3999999, with the exception of 4999999