There are a couple of things to do next while you're getting set up: making your site private and inviting your team.

If you've got a publication that you don't want the world to see yet because it's not ready to launch, you can hide your Ghost site behind a basic shared pass-phrase.

You can toggle this preference on at the bottom of Ghost's General Settings:

Ghost will give you a short, randomly generated pass-phrase which you can share with anyone who needs access to the site while you're working on it. While this setting is enabled, all search engine optimisation features will be switched off to help keep your site under the radar.

Do remember though, this is not secure authentication. You shouldn't rely on this feature for protecting important private data. It's just a simple, shared pass-phrase for some very basic privacy.

Ghost has a number of different user roles for your team:

Contributors
This is the base user level in Ghost. Contributors can create and edit their own draft posts, but they are unable to edit drafts of others or publish posts. Contributors are untrusted users with the most basic access to your publication.

Authors
Authors are the 2nd user level in Ghost. Authors can write, edit  and publish their own posts. Authors are trusted users. If you don't trust users to be allowed to publish their own posts, they should be set as Contributors.

Editors
Editors are the 3rd user level in Ghost. Editors can do everything that an Author can do, but they can also edit and publish the posts of others - as well as their own. Editors can also invite new Contributors & Authors to the site.

The top user level in Ghost is Administrator. Again, administrators can do everything that Authors and Editors can do, but they can also edit all site settings and data, not just content. Additionally, administrators have full access to invite, manage or remove any other user of the site.

The Owner
There is only ever one owner of a Ghost site. The owner is a special user which has all the same permissions as an Administrator, but with two exceptions: The Owner can never be deleted. And in some circumstances the owner will have access to additional special settings if applicable. For example: billing details, if using Ghost(Pro).

It's a good idea to ask all of your users to fill out their user profiles, including bio and social links. These will populate rich structured data for posts and generally create more opportunities for themes to fully populate their design.

## Next: Organising content

Find out how to organise your content with sensible tags and authors, or for more advanced configurations, how to create custom content structures using dynamic routing.

We’re releasing an analysis showing that since 2012 the amount of compute needed to train a neural net to the same performance on ImageNet 1 classification has been decreasing by a factor of 2 every 16 months. Compared to 2012, it now takes 44 times less compute to train a neural network to the level of AlexNet 2 (by contrast, Moore’s Law 3 would yield an 11x cost improvement over this period). Our results suggest that for AI tasks with high levels of recent investment, algorithmic progress has yielded more gains than classical hardware efficiency.

Algorithmic improvement is a key factor driving the advance of AI. It’s important to search for measures that shed light on overall algorithmic progress, even though it’s harder than measuring such trends in compute.

##### 44x less compute required to get to AlexNet performance 7 years later
Total amount of compute in teraflops/s-days used to train to AlexNet level performance. Lowest compute points at any given time shown in blue, all points measured shown in gray.25678910111213141516

## Measuring efficiency

Algorithmic efficiency can be defined as reducing the compute needed to train a specific capability. Efﬁciency is the primary way we measure algorithmic progress on classic computer science problems like sorting. Efficiency gains on traditional problems like sorting are more straightforward to measure than in ML because they have a clearer measure of task difficulty. [1]

However, we can apply the efficiency lens to machine learning by holding performance constant. Efficiency trends can be compared across domains like DNA sequencing17 (10-month doubling), solar energy18 (6-year doubling), and transistor density3 (2-year doubling).

We are standardizing OpenAI’s deep learning framework on PyTorch. In the past, we implemented projects in many frameworks depending on their relative strengths. We’ve now chosen to standardize to make it easier for our team to create and share optimized implementations of our models.

$pip install procgen # install$ python -m procgen.interactive --env-name starpilot # human
\$ python <<EOF # random AI agent
import gym
env = gym.make('procgen:procgen-coinrun-v0')
obs = env.reset()
while True:
obs, rew, done, info = env.step(env.action_space.sample())
env.render()
if done:
break
EOF


### Design principles

We’ve designed all Procgen environments to satisfy the following criteria:

• High Diversity: Environment generation logic is given maximal freedom, subject to basic design constraints. The diversity in the resulting level distributions presents agents with meaningful generalization challenges.

• Fast Evaluation: Environment difficulty is calibrated such that baseline agents make significant progress after training for 200M timesteps. Moreover, the environments are optimized to perform thousands of steps per second on a single CPU core, enabling a fast experimental pipeline.

• Tunable Difficulty: All environments support two well-calibrated difficulty settings: easy and hard. While we report results using the hard difficulty setting, we make the easy difficulty setting available for those with limited access to compute power. Easy environments require approximately an eighth of the resources to train.

• Emphasis on Visual Recognition and Motor Control: In keeping with precedent, environments mimic the style of many Atari and Gym Retro games. Performing well primarily depends on identifying key assets in the observation space and enacting appropriate low level motor responses.

OpenAI builds free software for training,
benchmarking, and experimenting with AI.