Hallucinations

Twitter

Facebook

Copy Link

Key Facts

Key CausesTraining Data Gaps, Overconfidence

Key DefenseRAG, Grounding, Fact-Checking

Top News

Nimble Raises $47M to Give AI Agents Access to Real-Time, Verified Web Data

Nimble just closed a $47 million Series B round led by Norwest to expand its agentic web search platform. The startup uses AI agents to scrape, validate, and structure live web data into queryable tables, solving a critical bottleneck for enterprise AI deployments. With over 100 customers including Fortune 500 companies, Nimble claims 99% accuracy in delivering trusted web data that integrates directly into existing data warehouses.

Technology3 sources

Nimble Raises $47M to Give AI Agents Access to Real-Time, Verified Web Data

Technology3 sources

West Midlands Police Copilot scandal exposes AI hallucinations in UK policing decisions

A UK Parliament committee found West Midlands Police relied on AI-generated hallucinations about a fictional football match to ban Israeli fans, exposing critical gaps in AI governance. Despite this, 21 police forces still use Copilot with no coordinated oversight, raising urgent questions about responsible implementation of AI in public order decisions.

Policy2 sources

West Midlands Police Copilot scandal exposes AI hallucinations in UK policing decisions

Policy2 sources

MIT Researchers Expose Hidden Biases and Personalities in Large Language Models

Researchers from MIT and UC San Diego have developed a method to identify and manipulate over 500 hidden concepts within large language models, including biases, personalities, and moods. The technique uses a Recursive Feature Machine algorithm to steer AI responses, improving performance while also exposing potential vulnerabilities like jailbreaking and hallucinations.

Science3 sources

MIT Researchers Expose Hidden Biases and Personalities in Large Language Models

Science3 sources

Intel scales back phone support, launches AI-powered assistant built on Microsoft Copilot Studio

Intel has launched Ask Intel, an AI customer support system built on Microsoft Copilot Studio, marking a major shift in how the semiconductor giant handles customer service. The move comes as Intel removes public phone support numbers in most countries and consolidates engagement around web-based systems, raising questions about accuracy and the future of human support agents.

Technology4 sources

Intel scales back phone support, launches AI-powered assistant built on Microsoft Copilot Studio

Technology4 sources

Horror Demon Photo Appears in Washington, D.C. Real Estate Listing After AI Goes Wrong

A Washington, D.C. rental property listing shocked house hunters when AI-generated images produced a horrifying demon photo emerging from a bathroom mirror. The bizarre AI errors highlight growing concerns about misleading images in real estate as 70% of realtors admit to using AI tools, raising questions about transparency and accuracy in property photos.

Entertainment3 sources

Horror Demon Photo Appears in Washington, D.C. Real Estate Listing After AI Goes Wrong

Entertainment3 sources

Alpha School's AI-powered education model faces scrutiny over faulty lessons and data concerns

Alpha School, endorsed by the Department of Education, promises to teach K-12 students in just two hours daily using AI tutors. But leaked internal documents reveal the AI-powered school generates flawed lesson plans that employees say do "more harm than good." Former staff describe students as "guinea pigs" while raising concerns about constant student monitoring and inadequate data security.

Technology2 sources

Alpha School's AI-powered education model faces scrutiny over faulty lessons and data concerns

Technology2 sources

Godot Game Engine Maintainers Struggle With Flood of AI Slop Pull Requests

The Godot game engine faces a crisis as maintainers drown in AI-generated code contributions. Rémi Verschelde warns the open source project can't sustain the onslaught of low-quality pull requests, with over 4,600 currently open. GitHub's AI promotion compounds the problem while maintainers question every new contribution's authenticity.

Technology4 sources

Godot Game Engine Maintainers Struggle With Flood of AI Slop Pull Requests

Technology4 sources

Anthropic releases Claude Sonnet 4.6 with human-level computer use and coding improvements

Anthropic launched Claude Sonnet 4.6, bringing frontier-level AI capabilities to free and Pro users. The update delivers improved coding skills that rival its premium Opus model, human-level performance in computer use tasks, and a 1 million token context window. Developers prefer the new model 70% of the time over its predecessor, with major gains in security against prompt injection attacks.

Technology18 sources

Anthropic releases Claude Sonnet 4.6 with human-level computer use and coding improvements

Technology18 sources

KPMG partner fined AU$10,000 for using AI to cheat on internal AI training exam

A KPMG partner in Australia was fined AU$10,000 for using artificial intelligence to cheat during an internal training course on AI. The unnamed partner uploaded training materials to an AI platform to help generate responses. More than two dozen staff at KPMG Australia have been caught cheating on internal exams since July, raising concerns about AI misuse in professional settings across major consultancy and accounting firms.

Business2 sources

KPMG partner fined AU$10,000 for using AI to cheat on internal AI training exam

Business2 sources

Anthropic and Pentagon clash over AI safeguards as $200 million contract hangs in balance

The Pentagon is pushing Anthropic to allow military use of Claude AI for all lawful purposes, but the company refuses to budge on restrictions around autonomous weapons and mass domestic surveillance. With a $200 million contract at stake, the Defense Department threatens to end the partnership while other AI companies show more flexibility.

Policy26 sources

Anthropic and Pentagon clash over AI safeguards as $200 million contract hangs in balance

Policy26 sources

Meridian raises $17 million to build auditable AI agents for financial modeling

Meridian emerges from stealth with $17 million in seed funding to tackle AI-powered financial modeling. The startup takes a standalone IDE approach rather than integrating into Excel, aiming to make AI agents more auditable and deterministic for regulated financial institutions. The company secured $5 million in contracts in December alone.

Startups2 sources

Meridian raises $17 million to build auditable AI agents for financial modeling

Startups2 sources

SwitchBot AI Art Frame turns text prompts into wall art, but AI gaffes reveal limitations

SwitchBot launched its AI Art Frame with E Ink display technology and AI image generation powered by Google's Nano Banana model. Available in three sizes from $149 to $1,299, the cord-free display promises two-year battery life but requires a $4 monthly subscription for AI features. Early reviews reveal the frame struggles with spelling errors and copyrighted content while offering limited photo management compared to traditional digital frames.

Technology2 sources

SwitchBot AI Art Frame turns text prompts into wall art, but AI gaffes reveal limitations

Technology2 sources

AI chatbots fail to improve medical advice for patients, Oxford study reveals

A University of Oxford study published in Nature Medicine found that AI chatbots offer no advantage over internet searches when patients seek medical advice. Despite large language models achieving 94.9% accuracy in controlled tests, real-world human-AI interaction in healthcare revealed a troubling gap between AI potential and performance, with patients struggling to provide complete information and receiving inconsistent advice.

Health14 sources

AI chatbots fail to improve medical advice for patients, Oxford study reveals

Health14 sources

AI in surgery faces scrutiny as TruDi Navigation System linked to strokes and patient injuries

A Johnson & Johnson surgical device saw malfunctions surge from 7 to over 100 after adding AI, with at least 10 patients injured. Two stroke victims have filed lawsuits alleging the TruDi Navigation System's AI contributed to botched surgeries, including skull punctures and arterial injuries. The incidents highlight growing concerns about inadequate safety standards as the FDA approves AI-powered medical devices at an accelerating pace.

Health3 sources

AI in surgery faces scrutiny as TruDi Navigation System linked to strokes and patient injuries

Health3 sources

X tests AI-powered collaborative notes to speed up Community Notes fact-checking process

X is testing a collaborative notes experiment that lets Community Notes contributors request AI-generated drafts from Grok. The feature aims to address criticism about slow fact-checking by using community feedback to refine AI-written notes. Only top writers can access the experimental feature initially, with plans to expand availability as the system learns from contributor ratings and suggestions.

Technology2 sources

X tests AI-powered collaborative notes to speed up Community Notes fact-checking process

Technology2 sources

Prev 1…91011…17 Next