SAFETY - AI Dynamics

Nolan’s AI Film: Exploring Musk, Altman, and Zuckerberg’s Visions

By

–

25 July 2023 13h11

It would be fascinating if Christopher Nolan made a movie about future AI and compared three different scenarios from @elonmusk
, @sama
, and Zuckerberg. I still don't understand what's on their minds, especially WorldCoin orbs…

→ View original post on X — @ki_young_ju

25 July 2023

OpenAI Generated Harmful Content Examples to Train AI Safety

By

@_karenhao

–

25 July 2023 11h29

3/ To have enough examples of the harmful content it wanted its AI systems to avoid, OpenAI researchers generated some with AI. In one case, they asked AI to write a post of a teenage girl whose friend had enacted self-harm. OpenAI discusses this here: https://
arxiv.org/pdf/2208.03274
.pdf
…

→ View original post on X — @_karenhao

25 July 2023

The Dark Side of AI Content Moderation

By

@_karenhao

–

25 July 2023 11h29

Imagine how much worse it becomes when the content is also violent & grotesque. This is unavoidable with generative AI. “It’s something that needs to get done,” said Mark Sears, CEO of @theCloudFactory
, which doesn't take on moderation projects. “It’s just so unbelievably ugly.”

→ View original post on X — @_karenhao

25 July 2023

AI Industry and Researchers Warn of Extinction Risk

By

@tamaramccleary

–

25 July 2023 9h10

#AI industry and researchers sign statement warning of ‘extinction’ risk https://
cnn.it/3C0kqBB #ethics #FutureofWork

→ View original post on X — @tamaramccleary

25 July 2023

RényiTester: Variational Approach to Differential Privacy Testing

By

@googleai

–

25 July 2023 0h26

At @icmlconf
? Today at the @_LXAI workshop at 1:15 PM, listen to Monica Ribero’s keynote speech and learn about RényiTester, a variational approach to testing and auditing algorithms for #DifferentialPrivacy. Learn more: https://
latinxinai.org/icml-2023

→ View original post on X — @googleai

25 July 2023

MIT Algorithm Protects Sensitive Data in Machine Learning Models

By

@mit_csail

–

24 July 2023 18h00

Sensitive data encoded within ML models can be extracted by malicious agents. MIT algorithm enables users to potentially add the smallest amount of noise possible to ensure that this data is protected: https://
bit.ly/3NKm3ss

→ View original post on X — @mit_csail

24 July 2023

Flynn warns AI capabilities are both impressive and dangerous

By

@sassoftware

–

24 July 2023 17h31

Flynn adds, on GAI: "This type of digital push button is simultaneously impressive and scary."

→ View original post on X — @sassoftware

24 July 2023

CoDoC: Human-AI Collaboration Framework for Medical Settings

By

@demishassabis

–

24 July 2023 16h19

Researchers from @googledeepmind and @googleAI have published CoDoC, which explores how we could harness human-AI collaboration in hypothetical medical settings to deliver the best results

→ View original post on X — @demishassabis

24 July 2023