The Fascinating development of AI: From ChatGPT and DALL-E to Deepfakes Part 1

Did this happen all at once? Or is this a culmination of years of research?

Dec 18, 2022

Introduction

The phrase “Artificial Intelligence” was coined John McCarthy in his 1955 Proposal titled “A PROPOSAL FOR THE DARTMOUTH SUMMER RESEARCH PROJECT ON ARTIFICIAL INTELLIGENCE”, but the concept of an artificial being with intelligence has existed for much longer. This is not to be confused with “Artificial General Intelligence” which describes an AI that exhibits a human level of intelligence over a wide variety of tasks. Think your I, Robots and Bladerunners. With ChatGPT crossing 1 million users in five days, it’s time to take a step back and examine how we got here.

Greg Brockman @gdb

ChatGPT just crossed 1 million users; it's been 5 days since launch.

Sam Altman @sama

little openai update: gpt-3, github copilot, and dall-e each have more than 1 million signups! took gpt-3 ~24 months to get there, copilot i think around 6 months, and dall-e only 2.5 months.

Deep Fakes

On July 11th 2017 The University of Washington posted on their news website a post titled Lip-syncing Obama: New tools turn audio clips into realistic video. In it, it linked to a paper to be presented at SIGGRAPH 2017 in August. The paper outlined a technique that was developed by University of Washington researchers. In it they say

Given the mouth shape at each time instant, we synthesize high quality mouth texture, and composite it with proper 3D pose matching to change what he appears to be saying in a target video to match the input audio track.

In the viral video we see versions of President Barack Obama speaking dialogue he had spoken in the past, in entirely new videos.

In the paper they describe how they did it.

Trained on many hours of his weekly address footage, a recurrent neural network learns the mapping from raw audio features to mouth shapes

But this was a continuation of a story that actually began two years prior.In 2015 the same group of researchers released a paper detailing how they were able to use pictures of an individual to create a 3D model, without the need to scan said individual.

In the video former President George Bush speaking. While he is speaking, various facial renders of celebrities and world leaders move in unison with him, lip syncing, and mimicking facial expressions. While this video did not get the same fanfare as the President Obama video two years later (82k views at the time of this article), it was a huge step forward in this technology from what had come before. Previously complex software and expensive equipment had to be used to generate the faces. Now it was being done with just pictures of an individuals face from multiple angles.

Since the general public has become aware of these technologies there has been an explosion of websites, apps and tools that use these technologies. One of the most notable examples is the porn industry. There are now websites were you can create deepfake videos of people in compromising sexual positions. Popular apps like Reface do faceswaps and animate pictures of user’s faces using AI and deepfake technologies. With the rise in these technologies, there is growing concerns about how they will be used in the future.

Ill intent

While this technology is not inherently bad, there is the potential to use this for nefarious reasons. WKRN reported on the rise in sextortion amongst teens, where teens are shown AI generated images of videos of themselves and blackmailed. Afraid no one will believe that images and videos are not real, they pay ransoms to have the picture deleted. In a landmark case, A Chalfont mom was accused of potentially using deepfake technologies to harass her daughters rivals on the cheerleading team. While later the prosecution said this in regards to the allegation

The Bucks County District Attorney's Office released a statement acknowledging that "deepfake" technology may not have been used in the case.

It still points to a future where deepfakes can be used to hurt individuals credibility and destroy their careers.

Hope for the future

There have been many efforts to detect deepfakes as the technology has become more prominent. Intel released an article on November 14, 2022 indicating that they had a technology that could detect deepfake videos with up to 96% accuracy. Academia is also hard at work developing tools to detect deepfakes as well as analyzing deepfake detection models for their accuracy. Legislation is also working to catch up with countries like the United States introducing the DEEP FAKES Accountability Act and the U.K outlining a bill to criminalize deepfake porn without consent.

And while deepfakes can be used by people with ill intent, they can also be used to help strengthen messages. In David Beckham’s campaign against malaria these technologies were used to help lip sync his face to the various languages the campaign was broadcast in

Deepfakes could also be used in the education space to teach history using the faces of individuals from the past. Lastly deepfakes could be used to help get over past trauma in therapy sessions by talking to loved ones who are no longer alive.

Deepfake technology can be used in a wide variety of applications, from entertainment, to virtual reality and education. It remains to be seen who will win in the arms race between deepfakes and their detection. As well as what people will continue to do with this technology. It is my hope that we can use this technology responsibly, and ethically to enrich the lives of the current and future generations.

Stay Tuned for Part 2.

Thanks for making it to the end of my article. If you In part 2 we will go over DALL-E the battle over AI art.

IV. DALL-E 2

Introduce DALL-E 2 and explain how it is related to the original DALL-E model.
Discuss some of the key features and capabilities of DALL-E 2, such as its ability to generate images from text descriptions.
Provide examples of the kinds of images that DALL-E 2 can generate and some of the potential applications for this technology.

V. Conclusion

Summarize the main points of the blog post and provide some final thoughts on the future of AI and natural language processing.
Encourage readers to share their own thoughts and experiences with these technologies in the comments section.

Deus In Machina

The Fascinating development of AI: From ChatGPT and DALL-E to Deepfakes Part 1

Did this happen all at once? Or is this a culmination of years of research?

Introduction

Deep Fakes

Ill intent

Hope for the future

Stay Tuned for Part 2.

Links

Deep Fakes

GPT

https://cdn.openai.com/research-covers/language-unsupervised/language_understanding_paper.pdf