The shadows that follow the AI generative models

Grab the slides: slides.com/cheukting_ho/ai-shadows/

Hello I am Cheuk

• Open-Source contributor
  
  
   
• Organisers of community events
  
  
   
• PSF director and fellow
   
• Community manager at OpenSSF

• Part of the Linux Foundation
• Focus on Open Source Security
• Not for Profit
• Security newsletters and blog
• Free courses:
  • Secure Software Development Fundamentals
  • Securing Your Software Supply Chain with Sigstore
• OpenSSF Day (OSS EU)
  • 18th Spet Bilbao

Warning:

This talk will involve topics that relate to sexual abuse and child sexual abuse

Can you name a generative AI model?

A lot of example

prompt

new content

model

What can we do with it?

https://www.forbes.com/sites/mattnovak/2023/03/26/that-viral-image-of-pope-francis-wearing-a-white-puffer-coat-is-totally-fake/

https://openai.com/research/jukebox

AI coding assitance

AI writing assistance in general

• Articles
• Project management
• Slides

Correctness

https://meta.stackoverflow.com/questions/421831/temporary-policy-generative-ai-e-g-chatgpt-is-banned?cb=1

New research (not peer-reviewed) from UCSD found that on GPT-4, 62% of the generated code contains API misuses

 

https://arxiv.org/pdf/2308.10335.pdf

We incorporated more human feedback, including feedback submitted by ChatGPT users, to improve GPT-4’s behavior.

https://openai.com/gpt-4

https://www.reddit.com/r/ChatGPT/comments/zqmy0u/chat_gpt_doesnt_like_to_disagree_with_you/

https://fortune.com/2023/07/19/chatgpt-accuracy-stanford-study/

A lot of example

prompt

new content

model

Prompt injection

Prompt injection attacks, as the name suggests, involve maliciously inserting prompts or requests in interactive systems to manipulate or deceive users, potentially leading to unintended actions or disclosure of sensitive information.

 

It’s similar to something like an SQL injection attack in that a command is embedded in something that seems like a normal input at the start.

https://hackaday.com/2023/05/19/prompt-injection-an-ai-targeted-attack/

SQL injection is still included in the Open Worldwide Application Security Project (OWASP) Top 10 list of security vulnerabilities.

https://www.darkreading.com/edge-articles/moveit-was-a-sql-injection-accident-waiting-to-happen

https://www.tomshardware.com/news/chatgpt-vulnerable-to-youtube-prompt-injection

Responsibility and ownership

https://www.nytimes.com/2022/09/02/technology/ai-artificial-intelligence-artists.html

AI-Generated Jokes

https://www.forbes.com/sites/danidiplacido/2023/02/06/ai-generated-seinfeld-banned-from-twitch-after-making-transphobic-jokes

Bias

https://twitter.com/Chicken3gg/status/1274314622447820801

https://twitter.com/ronawang/status/1679867848741765122

Credit: Anne Guérin

A lot of example

prompt

new content

model

Ethical Issues

More than

Noelle Martin was 17 when she discovered that her face had been edited onto naked photos of someone else...

her screen had been flooded by deepfake pornographic imagery –

featuring her face – created by an unknown group of “nameless, faceless” sexual predators.

https://www.dazeddigital.com/science-tech/article/55926/1/inside-the-disturbing-rise-of-deepfake-porn

Australia’s eSafety Commissioner has already received a number of complaints about non-consensual distribution of deepfake intimate images, and expects this type of abuse to grow in volume as artificial intelligence (AI) technology becomes more accessible.

https://cosmosmagazine.com/technology/ai/ai-automate-sexual-abuse-and-child-grooming/

What comes next?

Identity thieves?

Fake news?

Regulations

https://artificialintelligenceact.eu/

https://www.whitehouse.gov/ostp/ai-bill-of-rights/

AI Ethics Career

• Courses
• Jobs
• NGO
• Organization

https://training.linuxfoundation.org/training/ethics-in-ai-and-data-science-lfs112/

Thank you

Grab the slides: slides.com/cheukting_ho/ai-shadows/