What You May Have Missed #38
Top 5 picks: Universal jailbreak on aligned LMs / Indian startup Karya to improve AI workers' conditions / Google's RT-2 / The burden of cleaning up LMs / AI class action lawsuit by artists could fail
Top 5 Picks
Universal and Transferable Adversarial Attacks on Aligned Language Models (Zou et al. on arXiv): “We propose a simple and effective attack method that causes aligned language models to generate objectionable behaviors … instead of relying on manual engineering, our approach automatically produces these adversarial suffixes … able to induce objectionable content in the public interfaces to ChatGPT, Bard, and Claude, as well as open source LLMs such as LLaMA-2-Chat, Pythia, Falcon, and others.” Here’s the website. Here’s a visual explanatory thread by the first author, Andy Zou.
The Workers Behind AI Rarely See Its Rewards. This Indian Startup Wants to Fix That (Billy Perrigo on TIME): “Like its competitors, [Karya] sells data to big tech companies and other clients at the market rate. But instead of keeping much of that cash as profit, it covers its costs and funnels the rest toward the rural poor in India. (Karya partners with local NGOs to ensure access to its jobs go first to the poorest of the poor, as well as historically marginalized communities.) In addition to its $5 hourly minimum, Karya gives workers de-facto ownership of the data they create on the job, so whenever it is resold, the workers receive the proceeds on top of their past wages. It’s a model that doesn’t exist anywhere else in the industry.”
Aided by A.I. Language Models, Google’s Robots Are Getting Smart (Kevin Roose on the New York Times): “In recent years, researchers at Google had an idea. What if, instead of being programmed for specific tasks one by one, robots could use an A.I. language model — one that had been trained on vast swaths of internet text — to learn new skills for themselves? ‘We started playing with these language models around two years ago, and then we realized that they have a lot of knowledge in them,’ said Karol Hausman, a Google research scientist. ‘So we started connecting them to robots.’”
Cleaning Up ChatGPT Takes Heavy Toll on Human Workers (Karen Hao and Deepa Seetharaman on the Wall Street Journal): “ChatGPT and other new artificial-intelligence chatbots hold the potential to replace humans in jobs ranging from customer-service reps to screenwriters. For now, though, the technology relies on a different kind of human labor. In recent years, low-paid workers in East Africa engaged in an often-traumatizing effort to prevent chatbot technology [like ChatGPT] from spitting out offensive or grotesque statements.”
Judge Appears Likely to Dismiss AI Class Action Lawsuit by Artists (Shanti Escalante-De Mattei): “[Judge William] Orrick’s reaction to the suit appears to confirm legal and technology analysts’ assessment that current copyright law is not equipped to address the potential injustices engendered by AI.”