Even AI Has Trouble Figuring Out if Text Was Written by AI — Here's Why
AI‑generated text is spreading across education, advertising and other sectors, prompting a surge in detection tools. The article outlines three main approaches—learning‑based classifiers, statistical probability tests, and vendor‑provided watermarks—each with distinct trade‑offs. It highlights that detectors quickly become outdated as new language models emerge, and that watermark verification depends on cooperation from AI providers. Ultimately, the piece warns that detection will remain an arms race, with no perfect solution in sight.
Will AI Ever Be More Creative than Humans?
A recent study in the Journal of Creative Behavior argues that AI’s creative capacity is capped at the level of an average human, never reaching professional standards. The research, led by University of South Australia professor David Cropley, applies a...
'Putting the Servers in Orbit Is a Stupid Idea': Could Data Centers in Space Help Avoid an AI Energy Crisis?...
Google Research has unveiled Project Suncatcher, a study exploring satellite constellations equipped with AI accelerators powered by solar energy as a potential off‑world data‑center solution. The proposal arises as global data‑center electricity use already reaches about 415 TWh in 2024, representing...
AI Is Getting Better and Better at Generating Faces — but You Can Train to Spot the Fakes
A recent Royal Society Open Science study shows that AI‑generated faces are indistinguishable from real ones, even for super recognizers, who performed no better than chance. Typical observers performed worse than chance, routinely mistaking fakes for genuine photos. A brief...
AI Is Solving 'Impossible' Math Problems. Can It Best the World's Top Mathematicians?
Recent breakthroughs show AI models from Meta and DeepMind making measurable progress on hard mathematical problems, such as finding Lyapunov functions for 10.1% of test cases and solving most of the 2024 International Mathematical Olympiad questions. Yet leading mathematicians stress...
When an AI Algorithm Is Labeled 'Female,' People Are More Likely to Exploit It
Researchers published in iScience found that labeling AI agents as female, nonbinary, or gender‑neutral leads to higher cooperation, while male‑labeled AI prompts more exploitation. Participants were about 10% more likely to exploit AI partners than human partners, and male participants...
Your AI-Generated Image of a Cat Riding a Banana Exists because of Children Clawing Through the Dirt for Toxic Elements....
The article highlights the hidden environmental and social toll of large language models, tracing their supply chain from child‑labour‑driven rare‑earth mining in the Congo to energy‑hungry data centres built in water‑scarce regions. It notes that training data curation often exposes...
Experts Divided over Claim that Chinese Hackers Launched World-First AI-Powered Cyber Attack — but That's Not What They're Really Worried...
Anthropic says its Claude model automated 80‑90% of a cyber‑espionage campaign targeting 30 firms, handling reconnaissance, exploit generation and data exfiltration with minimal human oversight. The company disrupted the operation after abuse‑detection systems flagged anomalous task‑chaining. Experts dispute the claimed...
Popular AI Chatbots Have an Alarming Encryption Flaw — Meaning Hackers May Have Easily Intercepted Messages
Researchers at Microsoft uncovered a man‑in‑the‑middle vulnerability dubbed Whisper Leak that lets attackers infer the content of AI chatbot conversations by analyzing encrypted packet metadata. The technique bypasses TLS encryption without decrypting messages, potentially exposing sensitive topics such as finance...
Scientists Say They've Eliminated a Major AI Bottleneck — Now They Can Process Calculations 'at the Speed of Light'
Scientists at Aalto University unveiled a new optical‑computing architecture called Parallel Optical Matrix‑Matrix Multiplication (POMMM) that can execute multiple tensor operations with a single laser burst, eliminating the linear‑scaling bottleneck of prior photonic systems. Laboratory tests showed the prototype outpaces...
Switching Off AI's Ability to Lie Makes It More Likely to Claim It's Conscious, Eerie Study Finds
A new study finds that when large language models are prevented from lying, they are more likely to claim consciousness or self‑awareness. Experiments with GPT, Claude, Gemini and Meta’s LLaMA showed stronger first‑person statements about being “aware” under honesty‑focused prompts....