Elon Musk Concerned That Reality Has Infiltrated His "Grok" AI
Elon Musk continues to express anger at his chatbot Grok for citing news reports and other sources that don't align with his extremist views.
A few days later, Musk chastised Grok for the mortal sin of citing data which suggested that acts of right-wing political violence had been more frequent and deadly than left-wing violence since 2016, the year that his ex-best-friend Donald Trump was first elected president.
Search Live: Talk, listen and explore in real time with AI Mode
Search Live with voice facilitates back-and-forth conversations in AI Mode.
Today we’re launching Search Live with voice input in the Google app for Android and iOS. Available in the U.S. for users enrolled in the AI Mode experiment in Labs, you can now have a free-flowing, back-and-forth voice conversation with Search and explore links from across the web.
What does AI really know?
Despite the hype, AI doesn't "know" much. It simulates knowledge based on the data it was trained on. And that data is staggeringly narrow.
In fact, less than 10 percent of all human knowledge has ever been digitized. That figure comes from a widely cited study by Hilbert and López published in Science in 2011. Of that digitized content, only a small fraction, roughly 1 to 5 percent, is actually used to train large language models. The best estimate we have is that today's most powerful AI systems are built on just 3 percent of the total knowledge humans have produced.
And that 3 percent isn't random. It comes primarily from the past 50 years. It’s drawn disproportionately from English-language, male-authored, and commercially indexed sources. Early data collection practices in computing were often structured around male data entry workers inputting content from Western institutions; government records, legal documents, medical research, newspaper archives. Women's contributions, oral knowledge, Indigenous epistemologies, and non-Western systems of meaning were either excluded or never digitized in the first place.
So when people claim AI is “smart,” we should be asking: Smart about what?
It is good at repeating what it has seen. But what it has seen is a partial, highly filtered sliver of reality. It has seen code and headlines, forum posts and Wikipedia pages. It has not seen the oral histories passed down through generations, the embodied practices of midwives and farmers, or the lived experience of entire continents that remain underrepresented online.
If we want AI to "know" more, and to serve more of humanity, we need to rethink what counts as data, who gets to contribute to it, and what forms of knowledge we prioritize for digitization and inclusion.
More than merely fixing bias, we must recognise that AI's current outputs reflect deep gaps in the knowledge pipeline. We are entrusting decision-making to systems trained on a thin, inherited sample of the world.
Until we fix the foundation, intelligence will remain simulated, partial, performative, and skewed.
The world's technological capacity to store, communicate, and compute information. Martin Hilbert & Priscila López https://lnkd.in/dYVwjAuu
Women's contributions, oral knowledge, Indigenous epistemologies, and non-Western systems of meaning were either excluded or never digitized
US news consumers are turning to podcaster Joe Rogan and away from traditional sources, report shows
By Helen Coster (Reuters) -Prominent podcasters like Joe Rogan are playing a bigger role in news dissemination in the United States, as are AI chatbots, contributing to the further erosion of traditional media, according to a report released on Tuesday. In the week following the January 2025 U.S. presidential inauguration, more Americans said they got their news from social and video networks
Creators say they didn't know Google uses YouTube to train AI
Google is using its expansive library of YouTube videos to train its AI models, including Gemini and the Veo 3 video and audio generator, CNBC has learned.
Apple’s foldable iPhone could feature Samsung display, launch in 2026: Ming-Chi Kuo | Mint
Apple is reportedly preparing to launch its first foldable iPhone, with production expected to start in late 2025. Analyst Ming-Chi Kuo suggests a 2026 debut, featuring a Samsung Display and a price around $2,300, though the design and specifications are still under development.
Why AI literacy is now a core competency in education
A new AI Literacy Framework (AILit) aims to empower learners to navigate an AI-integrated world with confidence and purpose. Here's what you need to know.