• Welcome to the new Internet Infidels Discussion Board, formerly Talk Freethought.

Amazing new AI system - GPT-3 and generating images from text

OpenAI upgrades GPT-3, stunning with rhyming poetry and lyrics | Ars Technica - "Refinement to AI language model generates rhyming compositions in various styles."

Ethan Mollick on Twitter: "I posted about how image-generating AI has gotten exponentially better in the last month.
Well, a new text model was released for GPT-3 today.
AI can now write rhyming poems.
And acrostics.
And limericks.
And explain how a candy-powered FTL drive can help me escape from otters. (pix link)" / Twitter

The poem:
write me an rhyming poem about why I should love AI

AI is here to stay,
And it's here to play
It can help with your work,
And make it less of a jerk

It can answer your questions,
And help with suggestions
It can do things that you can't,
And help in a jiffy, it's no joke or stunt

AI can be of use,
And help you find the clues
It can make your life easier,
And be a helpful teaser

AI can help with your life,
It can be like a helpful wife
It can help you get more done,
And be a helpful ton

So don't be scared,
And embrace the AI instead
It can help you in ways unseen,
And make your life more serene
Click to expand...
The acrostic:
write me an acrostic about why I should be scared of AI

A - Artificial intelligence has the potential to create chaos
I - Instilling fear in the human race

S - Self-learning systems could lead to mistakes
C - Computers are not perfect and can malfunction
A - Autonomous systems have the power to cause destruction
R - Robots could be programmed to act against us
E - Excessive reliance on AI could lead to disaster
Click to expand...
The limerick:
write me a limerick about otters

An Otter from the river deep
Loved to swim and dive and leap
But when it came to rest
It chose a cozy nest
Inside a hollow log to sleep
Click to expand...
The explanation:
Explain to me in detail how my candy-powered Faster Than Light Drive can help me escape an angry otter

The candy-powered Faster Than Light Drive is a revolutionary invention that allows you to travel faster than the speed of light. This type of drive works by converting candy into energy, which is then used to propel the vehicle through space. By using this drive, you can escape an angry otter quickly and easily.

To use the candy-powered Faster Than Light Drive, you need to feed it with a variety of candies. Different types of candies will produce different levels of energy, so you need to find the right combination to power your vehicle. you have filled the drive with the right amount of candy, it will begin to convert the candy into energy. This energy will then be used to create a powerful force field that will propel the vehicle forward at speeds faster than light.

The Faster Than Light Drive is a great way to escape an angry otter because it allows you to move quickly and escape the situation before the otter can catch up. It is also a great way to travel long distances in a short amount of time, allowing you to reach your destination much faster than you would be able to without the drive.
Click to expand...
 
Write a short rhyming poem explaining Einstein's theory of general relativity in... | Hacker News
Write a short rhyming poem explaining Einstein's theory of general relativity in easy but accurate terms

If you want to understand Einstein's thought
It's not that hard if you give it a shot
General Relativity is the name of the game
Where space and time cannot remain the same
Mass affects the curvature of space
Which affects the flow of time's race
An object's motion will be affected
By the distortion that is detected
The closer you are to a large mass
The slower time will seem to pass
The farther away you may be
Time will speed up for you to see

first try with that prompt... yeah davinci-3 is awesome
btw "If" was also part of the prompt
Click to expand...
and
I prompted the model to re-write the poem in Keats' style by adding: "How would ... | Hacker News
I prompted the model to re-write the poem in Keats' style by adding: "How would Keats write this?"

Output:

Oh, what a wondrous force of might
That Einstein's mind did conjure up
To show us how the fabric of space
Is so profoundly warp and sup
Mass doth affect the shape of time
And cause our clocks to slow or speed
For time and space are linked together
As our great minds have here decreed
Click to expand...
 
Back to the article.
OpenAI makes GPT-3 available as a commercial product with an API, but for a fee ($0.02 per 1,000 tokens), anyone with an OpenAI account can experiment with the AI through a special "Playground" website that requires no coding skill.
Click to expand...
With examples
Write a poem about Ars Technica

Ars Technica, what a sight
A source of information so bright,
Filled with articles day and night
To tech fans, it's a delight.

From hardware to software,
From Mac to Windows,
Nothing escapes the eyes
Of Ars Technica's wise.

From the latest news
To the greatest reviews,
It's the place to go
To find out what's true.
Click to expand...
and
Write rhyming instructions on how to brew a cup of tea in the form of a poem:

Put your kettle on the stove
To boil some water for your tea
Add a teabag or two
o a cup for me and you

Steep for a few minutes
Then add a bit of sweet
Stir it all together
For a wonderful treat

Sip and enjoy your tea
Rising above the fray
Refreshing and delightful
On this lovely day
Click to expand...
and
Write a limerick about Benj Edwards being an AI

There once was an AI named Benj Edwards
Who focused on computer retro trends
He wrote and he spoke
On the past and its folk
He's the king of the tech history bends
Click to expand...
 
Introduced in 2020, GPT-3 gained renown for its ability to compose text in various styles at a similar level to a human, thanks to extensive training on text scraped from the Internet and data pulled from books. It uses statistical associations between learned word positions to predict the next best word in the sequence while reading from the prompt.
Click to expand...
noting openai/gpt-3: GPT-3: Language Models are Few-Shot Learners
noting [2005.14165] Language Models are Few-Shot Learners at ArXiv

Seems like GPT-3 is succeeding where  Cyc has failed.
Cyc (pronounced /ˈsaɪk/ SYKE) is a long-term artificial intelligence project that aims to assemble a comprehensive ontology and knowledge base that spans the basic concepts and rules about how the world works. Hoping to capture common sense knowledge, Cyc focuses on implicit knowledge that other AI platforms may take for granted. This is contrasted with facts one might find somewhere on the internet or retrieve via a search engine or Wikipedia. Cyc enables semantic reasoners to perform human-like reasoning and be less "brittle" when confronted with novel situations.

...
Much of the current work Cyc continues to be knowledge engineering, representing facts about the world by hand, and implementing efficient inference mechanisms on that knowledge. Increasingly, however, work at Cycorp involves giving the Cyc system the ability to communicate with end users in natural language, and to assist with the ongoing knowledge formation process via machine learning and natural-language understanding. Another large effort at Cycorp is building a suite of Cyc-powered ontological engineering tools to lower the bar to entry for individuals to contribute to, edit, browse, and query Cyc.
Click to expand...
Cyc is based on top-down AI, with explicitly specified inference rules, while GPT-3 is based on bottom-up AI, inferring inference rules from a training set of data.
 
The latest stuff:


This includes "Gaussian splatting" which is a 3D particle/cloud format that I think can even include transparent materials with refraction and reflection. It can get this data from photos and videos - or from AI as the video shows.
 
23 AI Tools You Won't Believe are Free - YouTube

More at Futurepedia - The Largest AI Tools & Software Directory
 
Another one: The Future of AI Video Has Arrived! (Stable Diffusion Video Tutorial/Walkthrough) - YouTube

At Futurepedia - The Largest AI Tools & Software Directory is AI Tools - Find the Best AI Software by Category with number of tools in each subcategory
  • text: copywriting 210, email assistant 142, paraphrasing 28, prompt generators 132, SEO 159, social media 231, story teller 82, summarizer 203, writing generators 294
  • image: art 146, avatars 109, design generators 223, image editing 166, image generators 266, image to image 29, logo generator 21, photoshop 29, text to image 28
  • code: code assistant 153, low-code/no-code 195, spreadsheets 60, SQL 51
  • audio: audio editing 62, music 107, text to speech 82, transcriber 126
  • video: text to video 27, video editing 79, video enhancer 30, video generators 120
  • 3D: 3D 70
  • business: customer support 220, e-commerce 68, education 259, fashion 20, finance 100, human resources 174, legal 40, presentations 39, productivity 523, real estate 32, sales 154, startup tools 247
  • other: fitness 33, fun tools 216, gaming 52, gift ideas 34, healthcare 90, religion 13, research 180, search engine 122, travel 68
 
For several months, I have been engaging in personal projects utilizing Realtime Voice Changer (RVC) to create unique music collaborations featuring renowned artists. This has been an incredibly enjoyable experience for me. Been using it for educational and personal amusement and don't plan on making any of it public.

Client
Voice Models

Advancements in AI have been fascinating to watch, particularly in the realm of voice changing technology. Currently, these tools are adept at altering voices, but they still require the user to mimic singing styles and mannerisms to achieve a convincing result. However, it won't be long before AI can independently replicate not just voices, but also singing abilities and mannerisms.


Edit: I realized I didn't clarify the connection to Chat GPT. Imagine if Chat GPT incorporated a Realtime Voice Changer (RVC) feature – the potential and capabilities of such an integration would be extraordinary. For example ask Chat GPT to write a song and have Kurt Cobain sing it in tenor at Tempo X. Then you can just play instruments to it.
 
The end of this video shows Tesla's Optimus Gen 2 moving an egg from one hand to another....


A few days ago Elon Musk tweeted:

So much engineering has gone into the hands. In a year, it will be able to thread a needle.
Click to expand...
That is mind blowing....

BTW I think Telsa's bots will be able to use a language model so that they can deal with natural language when people make requests.
 
Here's the latest from OpenAI a bit over 3 years after they released DALL-E.....

openai.com

Sora: Creating video from text

openai.com

Text to impressive high definition video... that link has dozens of examples including "Historical footage of California during the gold rush".

See: "Reminder: It’s only been 1 YEAR since the Will Smith eating spaghetti video" (which is horrible)



In the future I think AI would be able to generate video in real time and respond to keyboard/mouse input so you can interact with it.
 
Last edited:
So now hopeful new screen actors will have little chance of getting a job and established actors can stop working and just rent out their image.
 
excreationist said:
The end of this video shows Tesla's Optimus Gen 2 moving an egg from one hand to another....


A few days ago Elon Musk tweeted:

So much engineering has gone into the hands. In a year, it will be able to thread a needle.
Click to expand...
That is mind blowing....

BTW I think Telsa's bots will be able to use a language model so that they can deal with natural language when people make requests.
Click to expand...

Yeah, and Elon’s gonna send people to Mars and make a hyperloop for transit and have mental-telepathy brain chips and Yada.

In reality all he is good at is running Twitter into the abyss as a playground for right-wing turds like himself.
 
ZiprHead said:
So now hopeful new screen actors will have little chance of getting a job and established actors can stop working and just rent out their image.
Click to expand...
It could be difficult to always enforce the rental fee with videos like this - or they could just try and get the video removed....
 
Last edited:
Sora can also generate Minecraft gameplay videos.... at the moment to control the player you'd have to use the prompt to tell it what to do....
And of course it could mash up Minecraft with any other kind of conceivable visual style....
gizmodo.com

OpenAI’s Sora Is Coming for Your Video Games

The video generator excels at simulating digital worlds, and could completely disrupt the video game industry.
gizmodo.com gizmodo.com
2d3c4b0a1c18fa7ed5e05a037b9c4ffb.jpg
 
Sora as a rendering engine? That looks like an interesting possibility, but IMO that is impractical, since video-game rendering engines already exist. Sora seems like a generalized bottom-up renderer, and bottom-up won't work as well as top-down when one has an algorithm for what one wants to do.

But some Sora-like renderer may work for something difficult to model - one trains an AI engine on some things that take a lot of computer time to simulate, then uses that simulation if the AI engine can run fast enough.

I think that AI will do much better in game logic, where algorithms have much more limited success. This is evident in games with very simple game worlds, games like board games and card games, where AI is now at least as good as human players in the more difficult ones of them. For a video game, however, one may want to converse with NPC's, and one will need some good AI to do it well.

Some simple-world games have been solved:  Solved game - meaning that there is some strategy that enables at least one player to win or avoid losing. Tic-tac-toe (noughts and crosses, X's and O's) is a familiar one of these.

A larger game world means greater  Game complexity typically an exponential or factorial function of the game size. That makes that only games with very small game worlds have been solved, games like tic-tac-toe. The solved games with the largest game worlds are checkers and reversi.

Some games with larger game worlds are games where AI software can beat even the best human players, games like chess and Go.
 
One can make a taxonomy of game-world types, either deterministic (state entirely determined by the initial state and the players' actions) or stochastic (random element, like thrown dice or shuffled cards), and either complete-information (all the players can observe all of the game state) or incomplete-information (each player can observe only some of the game state).

Thus among board games and card games and the like,

Deterministic, complete-information: (solved) tic-tac-toe, checkers, (AI beats human players) chess, Go, ...

Deterministic, incomplete-information: battleship -- no work on AI play that I could find

Stochastic, complete-information: backgammon -- AI beats human players

Stochastic, incomplete-information: poker -- AI competitive with good human players
 
Game size is more difficult to define for continuous game worlds. Though they are ultimately discrete, that discreteness is very fine-grained, and defining a minimum effective size scale may be tricky. So I'll leave that aside.

I wish to note that 3D game worlds have had a progression over the decades.

Indoor:
  • Sidescrollers
  • First generation: same level of floor and ceiling, vertical walls on a square grid -- id's Wolfenstein 3D
  • Second ganeration: variable floor and ceiling levels, vertical walls with arbitrary horizontal locations -- id's Doom
  • Third generation: full-scale 3D models: arbitrary flat surfaces -- id's Quake
Outdoor:
  • Vertical or tilted flat surface
  • Variable-elevation terrain with sloped surfaces
Hybrid: indoor-engine buildings in an outdoor-engine landscape

Inhabitants:
  • Constant-sized sprites
  • Variable-sized sprites
  • 3D models
Sprites: image object

Color:
  • Indexed color: 2, 16, 256
  • Truecolor: 16-bit, 32-bit, more (?)
 
lpetrich said:
Sora as a rendering engine? That looks like an interesting possibility, but IMO that is impractical, since video-game rendering engines already exist.
Click to expand...
I'm not saying Sora would be used but an upcoming AI. Video games can take many years to develop to generate the characters and worlds. You could tell the AI the idea e.g. Las Vegas in the style of a Simpsons cartoon with zombies and you have a sword. Then you'd be able to play straight away.
BTW current rendering engines use polygons and shaders. Things like Sora don't.
 
openai.com

Reducing bias and improving safety in DALL·E 2

Today, we are implementing a new technique so that DALL·E generates images of people that more accurately reflect the diversity of the world’s population.
openai.com
People behind DALL-E were concerned that prompts for images such as firefighters were basically just generating white males.... then they fixed this.... (left side of images vs right side)
firefighters.JPG
Google's Gemini made the image results even more diverse:
qz.com

Google's AI was making really historically inaccurate pictures and now it's on ice

Google said its Gemini AI model was "missing the mark" with some images, after users shared pictures that included racially diverse Nazis
qz.com qz.com
GG0DNV4bQAAU7PF


"New game: Try to get Google Gemini to make an image of a Caucasian male. I have not been successful so far."
GG0cCCJXEAAn3aF
 
You must log in or register to reply here.
Back
Top Bottom