Let the Robot write your blog articles, programs and paint pictures (OpenAI - Part 1)

15 min read

Affiliate disclosure: This page contains affiliate links. Meaning, we get a commission if you purchase through our links, at no cost to you. (Read the full disclosure here.)

OpenAI is a private technology firm focused on artificial intelligence although they initially pivoted from deep learning. OpenAI was founded in December 2015, by SpaceX co-founder and Tesla CEO Elon Musk, Greg Brockman from notable data startup Cloudera, and entrepreneur Rebekah Mercer. Dimitry Ioffe, Pieter Abbeel, and Patrick Mynyk are also notable founding members of OpenAI. The company has partnerships with investment firms such as Fidelity Investments, Andreessen Horowitz), Obvious Ventures (led by Evan Williams of Twitter fame) and Ajay Chopra (ex-TCV). Ever since its creation OpenAI has made some large contributions to both its citizens and AI as a whole including developing new techniques for unsupervised learning & Natural language processing

One example is their work building safe AGI's (Artificial General Intelligence) which is where an agent or machine interacts autonomously with its environment striving to maximize some notion of cumulative reward. While this may sound like fancy scary terminology it’s actually our end goal for why we want intelligent machines in the first place so that they can help us achieve things faster than we could ever do alone.

Elon musk believes “If you're not concerned about AI safety then you should be very concerned because all civilization-ending events have been caused by technologies surging beyond our control." For this reason OpenAI established themselves with 3 initial goals:

1 Short term: To build advanced NLP algorithms that can read at least several pages of text

2 Midterm: Create robotic hand architecture capable of completing tasks like opening doorknobs

3 Long Term: To generate Artificial General Intelligence (AGI): Machines that equal or exceed human intelligence across the board rather than just narrow domains.

In 2016 They reached number one in world chess after creating their own computer program named AlphaGoZero. After only four days AlphaGoZero surpassed all previous versions Of Alphago (which had previously beaten the best human player Lee Sedol 4–1 ) as well as having taught itself how to play completely. What makes AlphaGoZero special compared to other programs Is it It’s completely self-taught unlike traditional chess programs which require access to thousands If not millions of professional grandmaster games in order to get good.

So what do you think of this article so far? It was written by OpenAI with just one instruction:

“write an article about OpenAI”

You can try it out yourself, head over to https://beta.openai.com/playground (you will have to sign up first), and type in the above command and see what it comes up with.

The OpenAI playground - let the Artificial Intelligence create a recipe from random ingredients.

Writing articles is not the only thing OpenAI can do for you. Here are some examples where OpenAI can be of use:

  • Question & Answer

  • Interview questions

  • Recipe creation (from a list of ingredients)

  • Create restaurant reviews

  • Essay outline

  • Grammar correction

  • Explain a complicated piece of code

  • Product name generator

  • Create SQL queries

What is OpenAI?

OpenAI is based on the GPT-3 (Generative Pre-trained Transformer 3), a language model that uses deep learning to produce human-like text. This language model was created by OpenAI, an artificial intelligence research laboratory in San Francisco. The model was trained by going through millions of different datasets, mostly websites (80%) and books (16%).

What can it be used for?

Besides the examples listed above, there is a wide range of use cases for GPT-3. It can be used for blog articles, sales emails, Facebook/Google ads, product listings, marketing copy, SEO content, headlines, and video scripts.

Jasper.ai, which is based on GPT-3 for example is a commercial tool that offers a set of features for bloggers, social media content, and websites.

Jasper.ai uses Copyscape to make sure that the content is plagiarism-free, they have Grammarly included, to improve your grammar and fix spelling mistakes and they use SurferSeo.com to make sure your content ranks high on search engines. Jasper.ai offers two modes: Starter and Boss Mode. The Starter Mode starts at 29$/mo with 20.000 words included and the Boss mode is 59$/mo with 50.000 words included. The more words per month you need the higher the monthly price.

We used OpenAI so far mostly for stock photo descriptions and tested it for blog articles. For short descriptions of pictures it works well and saves a lot of time, even if you don’t use all of the text that OpenAI produces, it is a good starting point or can give you some ideas for your writing.

When it comes to blog articles it needs a bit more editing and research. We asked OpenAI to write an article about a popular summer destination in the rocky mountains and it came up with some restaurant suggestions that were no longer around and hikes that were not even close to that region. So the results can’t just be used as is and needs to be reviewed carefully before being published.

The OpenAI playground reminds you on the side of the input field about these facts:

“Use good judgment when sharing outputs, and attribute them to your name or company. Learn more.

Most models' training data cuts off in October 2019, so they may not have knowledge of current events. “

This explains why it would still list a restaurant that has gone out of business, but it does not explain putting a popular hike in the wrong town.

Programming and painting

OpenAI can also do programming. We’ve tested OpenAI with a technical interview challenge for an application developer job. The coding challenge was a string calculator where you can enter a series of numbers (1,2,3,4,5) and the program adds the numbers.

The instruction for OpenAI could look like this:

“write a java function with streams including a test that can add the numbers from a string like this: 1,2,3,4,5”

And the result produced by OpenAI actually worked. But was it enough to impress the reviewers with the solution? Will they invite me for the next interview? To find out lookout for Part 2 of this article and subscribe to our newsletter.

Another interesting project of OpenAI is called DALL-E 2. With DALL-E 2 you can describe the contents of an image and the AI creates it for you. You can provide a text description like: “An astronaut riding a horse in a photorealistic style” and OpenAI turns it into a picture. To see the astronaut riding a horse check out the DALL-E 2 website: https://openai.com/dall-e-2/ .

DALL-E 2 is not yet available to the public. You can go to the following website and sign up for the waitlist for DALL-E 2: https://labs.openai.com/waitlist. We are on the waiting list and as soon as we can show you the first self-made masterpieces by DALL-E2 we will post them here.

Make sure to subscribe to our newsletter so you don’t miss any updates.

OpenAI Pros and Cons:

Pros:

  • Results for short text and short descriptions are very good

  • Length and temperature can be adjusted

  • Free

Cons:

  • Results for longer articles are often scrambled towards the end

  • Content not always accurate - facts needs to be checked carefully

In Part 2 about OpenAI we will dive deeper into the possibilities and limitations of OpenAI. With a little luck, we can show a preview of DALL-E 2 and we will find out if OpenAI was able to get me a job interview.

Links:

https://beta.openai.com/

https://www.jasper.ai/

https://www.grammarly.com/

https://openai.com/dall-e-2/

Previous
Previous

Plant a tree with every picture

Next
Next

Video recommendation week 18/2022