This Week in AI: ChatGPT Search Goes Live While Safety Concerns Mount | PYMNTS.com

OpenAI launched ChatGPT’s live web search to compete with Google this week, while the White House announced plans to treat artificial intelligence systems like controlled technology. Meanwhile, researchers discovered concerning accuracy problems with AI transcription tools, highlighting the gap between AI’s promise and its current limitations.

ChatGPT Gets Web-Savvy in Google Showdown

OpenAI just turbocharged ChatGPT with live web search capabilities, taking direct aim at Google’s search dominance. The updated AI chatbot, now rolling out to Plus and Team subscribers, can provide real-time info on everything from weather to stocks while citing its sources. Through partnerships with major news outlets like AP and Reuters, ChatGPT promises more accurate responses — potentially reshaping how we find information online.

AI Speech Tool’s Alarming Fabrications Raise Red Flags

OpenAI’s Whisper transcription software is making waves for all the wrong reasons, inventing dialogue that was never spoken — including disturbing content about violence and race. Research shows the AI tool hallucinates in 80% of transcripts, raising serious concerns as healthcare providers adopt the technology. The discovery comes as businesses pour billions into AI customer service, despite growing evidence of reliability issues.

Tiny Neural Networks Match Giants With Just 19 Neurons

MIT spinoff Liquid AI has demonstrated that neural networks barely larger than a worm’s brain can guide self-driving cars as effectively as systems using millions of neurons. The breakthrough “liquid networks” focus on essential data like road edges while ignoring distractions, potentially slashing the massive computing costs that have kept many businesses from adopting AI.

White House Treats AI Like Weapons in New Export Crackdown

The Biden administration is moving to classify artificial intelligence as a national security asset, requiring tech giants like OpenAI and Microsoft to get government approval before sharing innovations abroad. The directive orders federal agencies to shield private-sector AI breakthroughs like military tech from foreign access. Experts predict AI could soon face weapons-grade export restrictions, particularly for non-allied nations.

Google’s ‘Project Jarvis’ to Let AI Take Control of Your Browser

Google is reportedly developing an AI system that can independently operate web browsers, potentially transforming how we shop and surf online. The secretive “Project Jarvis,” set to launch alongside Google’s Gemini models in December, will navigate websites by interpreting screen content and executing actions like clicking and form-filling. While still slow, the technology signals a future where AI agents handle complex online tasks independently.