July 22, 2024

This Week in Responsible AI, Jul 22, 2024

Jailbreaks

• OpenAI’s latest model will block the ‘ignore all previous instructions’ loophole

• Refuse Whenever You Feel Unsafe: Improving Safety in LLMs via Decoupled Refusal Training

Privacy

• Trust No Bot: Discovering Personal Disclosures in Human-LLM Conversations in the Wild (PDF)

• The app that promised to ‘use AI to weed out daters with STIs’ has been shut down (CW: potentially NSFW)

Hype

• Everyone Is Judging AI by These Tests. But Experts Say They’re Close to Meaningless

• Large Models of What? Mistaking Engineering Achievements for Human Linguistic Agency

• Questionable practices in machine learning

• A Sanity Check on ‘Emergent Properties’ in Large Language Models

Social algorithms

• We unleashed Facebook and Instagram’s algorithms on blank accounts. They served up sexism and misogyny

• I Changed My Race to White on Hinge And got better matches. Is the algorithm the problem, or the men?

Law/Policy

• Meta will withhold multimodal AI models from the EU amid regulatory uncertainty

• OpenAI Dropped From First Ever AI Programming Copyright Lawsuit

• The AI Executive Order through the lens of the AI Index

• Biden’s top tech adviser says AI is a ‘today problem’

Copying

• Academic authors 'shocked' after Taylor & Francis sells access to their research to Microsoft AI

• Figma explains how its AI tool ripped off Apple’s design

• European Innovation Council: Artificial intelligence and copyright: use of generative AI tools to develop new content

• Disney’s internal Slack was leaked by hackers mad about AI

Other

• Want to spot a deepfake? Look for the stars in their eyes

• Data workers detail exploitation by tech industry in DAIR report

• AI and facial recognition tools in open-source intelligence

• Selfie-based authentication raises eyebrows among infosec experts

• AI AI Bias: Large Language Models Favor Their Own Generated Content

• Sustainable AI: a contradiction in terms?

Compiled by Leif Hancox-Li