Bias

An update of Google’s flagship multimodal model keeps track of colossal inputs, while an earlier version generated some questionable outputs.

Hugging Face introduced four leaderboards to rank the performance and trustworthiness of large language models (LLMs). The open source AI repository now ranks performance on tests of workplace utility, trust and safety, tendency to generate falsehoods, and reasoning.

In a study, models used to detect people walking on streets and sidewalks performed less well on adults with darker skin and children of all skin tones.

Amazon launched a chatbot for large companies even as internal tests indicated potential problems. Amazon introduced Q, an AI-powered assistant that enables employees to query documents and corporate systems.

An open source tool automatically tests language and tabular-data models for social biases and other common issues. Giskard is a software framework that evaluates models using a suite of heuristics and tests based on GPT-4.

How can we build large-scale language and vision models that don’t inherit social biases? Conventional wisdom suggests training on larger datasets, but research challenges this assumption.

Stable Diffusion may amplify biases in its training data in ways that promote deeply ingrained social stereotypes.

While Microsoft and Google scramble to supercharge their businesses with text generation, Meta has yet to launch a flagship generative AI service. Reporters went looking for reasons why.

The crowdworkers you hire to provide human data may use AI to produce it. Researchers at École Polytechnique Fédérale de Lausanne found that written material supplied by workers hired via Amazon Mechanical Turk showed signs of being generated by ChatGPT.

A new regulatory body created by the European Union promises to peer inside the black boxes that drive social media recommendations. The European Centre for Algorithmic Transparency (ECAT) will study the algorithms that identify, categorize...

Do language models have their own opinions about politically charged issues? Yes — and they probably don’t match yours. Shibani Santurkar and colleagues at Stanford compared opinion-poll responses of large language models with those of various human groups.

A pioneer of deep learning joined the chorus of AI insiders who worry that the technology is becoming dangerous, saying that part of him regrets his life’s work.

Meta’s effort to make a large language model available to researchers ended with its escape into the wild. Soon after Meta started accepting applications for developer access to LLaMA, a family of trained large language models...

AI hobbyists created an homage to their favorite TV show . . . until it got knocked off the server. The creators of Nothing, Forever launched a fully automated, never-ending emulation of the popular TV show Seinfeld.

The United States government published guidelines designed to help organizations limit harm from AI. The National Institute for Standards and Technology, which recommends technological standards in a variety of industries, released the initial version of its AI Risk Management Framework.