😲AI Hallucination Exterminator!?

Microsoft’s latest tool aims to ‘correct’ AI hallucinations, but experts remain skeptical

Key Points 

• Microsoft's new service, Correction, targets AI hallucinations by automatically fixing inaccuracies in generated text for improved reliability.

• The service identifies potentially incorrect text and fact-checks it against trusted sources.

• Experts warn that Correction may not address the root causes of AI hallucinations and could mislead users.

👨🏻‍💻News - Microsoft has rolled out a new service called Correction to deal with AI hallucinations, which is when AI spits out incorrect information. This service aims to automatically fix any inaccuracies in the text generated by AI, making it a lot more reliable.

🕵🏻‍♂️How does it work? Correction works by first identifying potentially incorrect text, such as a blog post about a movie that misstates who directed it. It then fact-checks that information by comparing it to trusted sources, like official movie databases

Currently, Correction is available in preview as part of Microsoft’s Azure AI Content Safety API. You can use it with any text-generating AI model, including Meta’s Llama and OpenAI’s GPT-4.

🧐What do experts think? Some experts are cautioning that methods like Microsoft’s don’t really tackle the root cause of AI hallucinations. Os Keyes, a PhD candidate at the University of Washington, put it bluntly: trying to get rid of hallucinations is like trying to take hydrogen out of water—it's just part of how the technology works. 

Mike Cook, a research fellow at Queen Mary University who studies AI, added that even if Correction does what it’s supposed to, it could create more trust and explainability problems. It might catch some errors, but it could also trick users into thinking AI is being more truthful than it really is.

World’s first AI art museum 'Dataland' to open in Los Angeles in 2025

👨🏻‍🎨News - Refik Anadol, a media artist known for his crowd-pleasing yet controversial AI-driven works, has announced that he will be opening the world’s first AI art museum, called Dataland, in Los Angeles.

Set to debut in late 2025, the museum will focus on the blend of human imagination and machine creativity. It will be located right next to the Museum of Contemporary Art, the Broad Museum, and the LA Philharmonic, making it a vibrant new spot for AI art among some of the city's top cultural attractions.

🖼️What's the vision? With Dataland, Anadol and his team of artists and tech experts want to reinvent the museum experience for the age of AI. They want to shine a light on the innovative work of digital artists, who have often faced skepticism from traditional art institutions, while also providing a space for ongoing scientific and technological research.

He’s really focused on promoting “ethical AI,” making sure to get permission for all sourced materials—something not everyone does in AI training. He also plans to power the museum with renewable energy. Plus, he intends to outfit the space with cloud computing, special sensors, and interactive activities to enhance the experience for visitors.

😁What can we expect to see? 

• Dataland is privately funded and will focus on collecting and preserving AI art, with some pieces possibly being sold on the blockchain.

• Anadol describes it as a “living museum,” built from pixels and voxels, which are just ways to represent 3D images mathematically.

• A major highlight is its own AI model, the Large Nature Model. Developed by Anadol’s studio, this model pulls data from several partners, including the Smithsonian, which has 9 million public specimen records and 6.3 million images; London’s Natural History Museum, with 90 million specimens and 4 million public images; and New York’s Cornell Lab of Ornithology, which offers 54 million images and 2 million sound recordings. With all this data, AI will create artworks based on up to half a billion images of nature.