Weird AI

The Open AI project announced DALL-E today which is something machine learning that something something creates images from text by something something. I have read the blog post announcing it a number of times and feel that I understand what they are saying. However I still don’t really understand any of what they are saying.

Machine Learning and Artificial Intelligence are these concepts that I am aware of. I understand ways in which they are being used, but I have no idea how they work. To some people this is scary and there are scenarios like the Terminator where AI is very scary. In some ways expecting AI to be as bad or worse than humans is logical.

The problem is that even with the advances that have been made recently we are still a long ways from a machine thinking anything close to a human mind. For example the DALL-E project is very bleeding edge stuff made by the smartest people on the planet. For their efforts they can give the computer text and it will make an image from that. Not quite a humanoid robot shooting at you.

From what I have gathered over the years of reading about the science of machine learning and AI is that it is going to be a gradual shift. We already have machine learning making our photos way better by auto adjusting ISO, Aperture, and Exposure. Automatic face detection for Portraits is done with Machine Learning.

These are helpful things but still limited. The phone is not recommending that you change the direction of the camera to capture better light, yet.

Last year OpenAI developed GMT-3, which can write somewhat coherent text, aka a langauge model. GMT-3 hit the headlines as journalists everywhere realized they might have to get jobs.

The reality is that Machine Learning is still in very early days and most writers have nothing to worry about, yet. Some humorous examples can be found on AI Weirdness such as new take on Rudolph the Red Nosed Reindeer:

All Hail to Rudolph

All hail to Rudolph
The legend of Christmas
The most famous reindeer of all

All hail to Rudolph
Happy all the year
From everyone we see

Heroes come and heroes go
But Rudolph stays forevermore
Oh boys and girls of every age
Can’t you hear the sleigh bells ringing?

Oh look! There’s Santa and Parson Brown
Defying the laws of time and space
Now prepare yourself and your family my boy
Prepare to get your Christmas fix

Now Rudolph goes smashing through the roof
Just like a power lift with a candy-cane moustache
And the deer’s nose is dead-on, deadly accurate
Pow! Now it’s Rudolph versus Mars

All hail this joyful night
All hail the night folks sing
All hail to Rudolph
The legend of Christmas
The most famous reindeer of all

O Come Rudolph, Come

O Come Rudolph, Come Ye Faithful
Oh come, come holly, clad in white, Oh come ye faithful.
Oh come aloud! Oh come and join our choir,
Oh make a joyful noise till earth and heaven ring.

Caw caw caw oh, caw caw caw oh, Caw holly, Caw plum this Christmas.
Oh, caw caw caw, oh caw oh, caw oh oh oh, oh.

Fun stuff but writers are safe at their jobs for at least a little longer. What is interesting is how we will be able to use these technologies to create. How could a writer or artist benefit from using these tools?

Brainstorming sessions might benefit training a model with your competitors websites/products/presentations etc. Things that we should already be doing but now you can literally just throw everything into a folder and let the software come up with new ideas.

Here is an example of a brainstorming meeting by Jimi Smoot where only the human inputs are in bold:


START EXPERIMENT:

The below is a conversation overheard in a product design meeting at a software company. The team has a Leader, two graphics designers, and a product manager. They are brainstorming ideas for a new software product to sell to Sales teams and help the teams close more deals.

Leader: OK remember we need to come up with something a sales team needs. Any ideas?

Designer 1: Why don’t we make an application where we record all of the steps a salesperson takes to close a deal.

Designer 2: Wow that sounds great. Anyone have any other ideas?

Product Manager: We should make an application that has all of the features you need to close a deal. Like a CRM with checkboxes to tick when you reach certain parts of the sales process.

Designer 2: Yes, that sounds great, I like your idea a lot. What else do we need?

Leader: What about a budgeting tool?

Designer 2: Yes that would be great.

Designer 1: OK that sounds good, we will put these ideas together and think about it some more. What about you guys? Do you have any more ideas?

Designer 3: Why don’t we make a checkbox application like the one the Product Manager mentioned. Sales people like checkboxes.

Designer 2: Yes, that sounds great, I like your idea a lot. What else do we need?

Leader: What about a budgeting tool?

Designer 3: Yes that would be great.

Leader: No I don’t like that one. Any other ideas?

Designer 2: Why don’t we make a checkbox application that has all of the features you need to close a deal. Like a CRM with checkboxes to tick when you reach certain parts of the sales process.

Leader: We already talked about that. How about something fresh? Are there any tools that can be used for selling with email?

Designer 3: I have an idea! What about we build an application that uses Gmail?

Designer 2: Yes that sounds great, I like your idea a lot. What else do we need?

Designer 1: We could do that. We need a way to connect to Gmail and work from Gmail. That way all of the sales person’s email is in one place.

Designer 2: OK we will think about that. What else do we need?

Product Manager: Let’s take a look at other companies that sell to sales teams. Are there any tools or applications that can be used for selling with email?

Leader: No one is going to buy a software product to help them sell with email. People will just use Gmail for that. Let’s look for something else.

Designer 2: Why don’t we make a checkbox application that has all of the features you need to close a deal. Like a CRM with checkboxes to tick when you reach certain parts of the sales process.

Leader: We already talked about that. Are there any tools that can be used for selling with email?

Product Manager: Why don’t we make an application where we record all of the steps a salesperson takes to close a deal.

END:


Again this stuff is early days but compared to some of the brainstorming sessions I have participated in this one is not bad.

David Bebbet kicks the tires of GPT-3 in a post from July of 2020: Writing with GPT-3. He shows a brainstorming session starting with 5 ideas from him then the machine gives him this:


And it came back with 15 more:

  1. Create a list of Tor commands to run to defend against an attack
  2. Create a cheat sheet for screenwriters
  3. Generate advice for a scene
  4. Generate XKCD comics
  5. Create a crib sheet for an arbitrary Python API
  6. Make a list of movies to watch
  7. Make a list of scholarly algorithms/heuristics worth reading
  8. Make a list of generic questions about Pedagogy
  9. Make a list of cognitive exercises
  10. Show patterns generated by awesome people
  11. Generate small cooking tasks
  12. Generate characters for a project
  13. Generate configuration options
  14. Generate flashcards
  15. Organize my computer/files

And then 15 more:

  1. Explain things to my parents via messenger
  2. Explain things to my children via messenger
  3. Explain mathematical algorithms
  4. Automated interface design
  5. Mechanize a Nagihara Kan Airlines website
  6. Clickbait generator
  7. Music description (Herbie Hancock is playing “Dolphin Dance” live in London on the 18th March 1968)
  8. Tell me not to waste my time looking at x
  9. Interesting facts about Wikipedia
  10. Explain something to someone while my attention is elsewhere
  11. Detect adverts in x
  12. Image caption/caption game
  13. Speculate about the true utility function of x
  14. Auto-generate tests for a program
  15. Explain why my code is currently failing tests.

And then 15 more:

  1. Generate plots or character/story summaries
  2. Generate legal examples
  3. Generate words for crosswords
  4. Check my grammar
  5. Supply filler text for themed writing (horror, sci-fi etc.)
  6. Explain why an AI can’t read (can it comprehend metaphors?)
  7. Provide entertaining answers to obscure questions
  8. Get writing ideas and generate word combinations
  9. Create background for an NPC I need for a D&D game
  10. Help a sick writer generate plots
  11. Input pre-generated (from books etc.) phrases and use them for literary analysis
  12. Help explain a mathematical proof
  13. Develop a game/story/character I need
  14. Make short stories I can “overlay” on a fictional Universe I’ve created (ala Ready Player One)
  15. Make a short film/audio drama

From Writing with GPT-3


David then teaches it “how to evaluate simple expressions and take derivatives”. My eyes got a little glossed over here to be honest. I haven’t seen any of those words since 1995.

Fortunately he creates a list of Overall Principles:

  1. GPT-3 is continuing your writing, not having a dialog with you.
  2. GPT-3 effectively puts your writing in a made up context.
  3. GPT-3 likes lists and repeated structures.
  4. GPT-3 cannot perform linguistic tricks like unscrambling words
  5. GPT-3 benefits from “showing its work”
  6. GPT-3 has a great breadth of knowledge and writing styles available to it
  7. Sometimes GPT-3 makes no sense

So it sounds like things are Weird and only going to get more so.

Leave a Reply