The AI Plug
Posts
#3 Deepfakes are nothing new

#3 Deepfakes are nothing new

AND: You won't believe these AI movies

The AI Plug
February 23, 2023

What do you think - Can AI create memes?

If you thought “no, AI can’t generate memes - that’s too human” - you’d have been in the majority - but believe it or not, you’re well behind the bell curve, because

AI has been generating memes for at least three years (since 2019 atleast).

This technology can be used for putting together what we think are some bloody funny memes (see below). However, it’s the same process that’s widely accessible for creating scarily-convincing deepfakes - a real problem in modern society.

In today’s newsletter:

Meme machine: Memes, deepfakes, and the internet is a vast wormhole of darkness that eats itself
The rundown: Zero-budget filmmaking, data talks, and robots training robots
Tips, tricks & tools: Words to sims
Zero-budget filmmaking: Animated film no longer cost $20,000 a minute
Monday job board: Career opportunities in AI

MEMES, DEEPFAKES, AND THE INTERNET IS “A VAST WORMHOLE OF DARKNESS THAT EATS ITSELF”

An AI-generated meme from This Meme Does Not Exist back in April 2020

Dylan Wenzlau, founder of Imgflip (the meme and GIF-creation site with 5.5 million monthly visits equalling a traffic value of $1.3 million 😮), created the “This Meme Does Not Exist” feature in 2019.

Using 100 million publically available meme captions to train his “This Meme Does Not Exist” AI, Dylan’s AI creates unique, and often gut-bustlingly funny captions for 48 of the site’s most popular meme templates.

While this might seem benign, like joking about Jerry in a helicopter, this process of combing templates with training data to build AI’s capable of superimposing images can be used for other, less noble, causes.

Like making deepfakes.

While we’re hearing more about the frankly jaw-dropping potential and consequences of deepfake technology by the day (think Tom Cruise doing magic tricks, or Snoop Dog giving tarot reads), the reality is, people have been deepfaking for years.

Back in 2018, one of the first major deepfake scandals hit the headlines, with Scarlet Johansson being deepfaked into multiple porn videos.

Even back then, Scarlet realized what we were up against, telling The Washington Post trying to stop deepfake porn was a “lost cause”, and that “the internet is a vast wormhole of darkness that eats itself.”

Fast forward 5 years to where we are today, and we’re seeing not only celebrities but everyday people copping the brunt of deepfake pornography.

The February 2023 Twitch star scandal highlights how convincing this technology is, and just how far behind regulation and laws are regarding the non-consensual manipulation of a person’s digital image.

What’s more, the potential application of deepfake technology to spread “fake news”, incite arguments between people, leaders, or states, or just deceive everyone until we don’t know up from down is very real.

The reality is, this technology is only going to continue developing, getting more widespread, and impacting more people. What’s perhaps most alarming is that there isn’t too much anyone can do about it.

Our advice? Approach anything you see online with well-measured scepticism, caution, and a pinch of cynicism. Remember, nothing is as it seems.

THE RUNDOWN 🐂

Zero-budget filmmaking is on the horizon, with Runway Research’s GEN-1 AI capable of generating new videos out of existing ones, based on word and image prompts.

Data Universe, a new global data & analytics event, will be hosted by RX US in New York, April 2024. The Data Universe event aims to bring professionals from all data disciplines together, from analytics, MLAI, strategy, operations and governance. Maybe they can come up with a way to guardrail the spread AI technology. Doubtful though…

Deep Learning Robotics (DLR)’s next-gen technology enables robotic devices to assist in the training of objects’ visual and 3D features to computer vision models. And now, we’ve got robots teaching robots. Which according to some, spells the end of humanity (more on this next week "*NOT ALARMIST JUST REALISTIC).

Mission-critical crisis scenarios now have unparalleled workflow automation and strategic briefing, with Blackbird’s Constellation Risk Engine-powered “RAV3N Copilot”. RAV3N Copilot represents “a major leap forward in the automated response capabilities that were previously unimaginable just a few years ago.”

Steve Jobs lives on in spirit in an AI-assisted product creation studio where professionals gather to work on product and content ideas using ChatGPT and imageAI. The studio, /imagine, aims to bridge the gap between in-person collaboration which is so easily replaced by software.

TIPS, TRICKS & TOOLS ⚒️

New-age storytelling

Normally here, we put together a list of AI tools which you can use to make your life better or your job easier.

Today, we bring you OPUS.

Why only a single tool, you ask?

Because, with this tool, you “write your stories, and OPUS will turn them into movies.”

Not sure about you, but our minds just got melted.

ANIMATED FILM NO LONGER COSTS $20,000 A MINUTE

Award-winning director Karen X. Cheng explores the new frontier of AI filmmaking

The industry standard for professionally-produced animated videos normally clocks in at around $10,000-$20,000 per minute.

A rather prohibitive barrier if happen to be an aspiring filmmaker who has a brilliant vision in mind, but no funding or investors ready to bring it to the big screen.

Don’t lose hope though.

In fact, get pretty excited.

Why?

Because watch Karen X. Cheng seriously, seriously cut down the budget required for filmmaking, using Runway Research’s GEN-1 Generative AI.

How?

By combining:

Video footage shot on a handphone
Image prompts with Midjourney-generated AI images
Text prompts

While the technology seems to be there, Karen clearly has an incredible skillset in imagining and designing mini sets, and also in prompting the software.

In fact, in her twitter thread, she says “each clip represents 50-100 tests and combinations - playing with the prompt, parameters, and reshooting footage.” (click below to follow her full thread, it’s definitely worth it).

AI is not a magic bullet - it still takes tons of human input and experimentation.
For every clip you saw, I did 50-100 tests and combinations - playing with the prompt, parameters, and reshooting footage.
There was definitely a learning curve, and my results starting looking… twitter.com/i/web/status/1…
— Karen X. Cheng (@karenxcheng)
Feb 20, 2023

BUT SERIOUSLY…

This newsletter has looked at some downright amazing/mind-blowing developments and applications in modern AI.

We touched on one story in The rundown though, where robots are starting to train AI systems.

And honestly?

That’s not good.

At least, not according to Eliezier Yudkowsky, an AI researcher who believes we’re, well, too late, and that we’re speeding towards impending AI-generated doom.

We’ll take a deep dive into that one in the next edition.