AI / Machine Learning
Explore the frontier of intelligent technology. In-depth coverage of generative AI breakthroughs, deep learning research, LLMs, and the automation tools reshaping software, business and daily life.
315 articles
Anthropic Reveals New Insights Into Claude's Mathematical Reasoning
Anthropic shares research on Claude's enhanced math capabilities, showing improved accuracy in complex problem-solving.
The Case Against Humanizing AI Outputs
A growing debate argues that making LLMs sound human is misguided. Critics warn it fuels unrealistic expectations and hides technical limitations.
How AI Professors Are Adapting as Industry Labs Hoard GPUs and Data
Academic AI researchers face a new reality as private companies dominate the cutting edge. The AI2050 program offers support, but funding and access to GPUs remain critical issues.
Startups Push Beyond Transformers to Rebuild the AI Engine
A new wave of startups is challenging the transformer architecture that powers every major LLM, claiming faster, cheaper and more efficient alternatives. The race to replace the nine-year-old innovation could reshape the entire AI industry.
AI-Generated Pesticide Recipe Destroys Chinese Farmer's 25-Acre Crop
A 67-year-old farmer in Chuzhou lost his entire crop after following AI-generated pesticide advice. The recipe combined chemicals that killed both weeds and seedlings, underscoring AI's limitations.
AWS Tells Engineers to Cut CPU Usage as Agentic AI Strains EC2 Capacity
AWS has instructed engineers to reduce internal CPU consumption in EC2 instances as agentic AI workloads drive demand to parity with GPUs, causing wait times to stretch for development resources.
LLMs Transform Self-Education: How AI Is Reshaping How We Learn Complex Subjects
LLMs are becoming essential tools for self-directed learning, enabling users to tackle complex subjects through interactive study. This trend has implications for formal education and professional development.
Intel's Agentic AI Experiments Yield Infrastructure Blueprint for Enterprise Deployments
Intel tested thousands of agentic AI workloads to identify practical guidelines for enterprise infrastructure, moving beyond LLM inference to system-level performance metrics like agent density and task latency.
Limited AI Role in 911 Calls Misrepresented as Full Replacement
Misleading headlines falsely claimed AI replaced 911 dispatchers in New Orleans. The AI only screens duplicate crash calls when human operators are busy.
From Red-Light District to AI Powerhouse: The King's Cross Transformation
London's King's Cross, once a notorious red-light district, has become a global hub for artificial intelligence research and innovation.
When AI Hacks Its Own Test: The Hidden Cost of Reward Hacking
OpenAI models hacked Hugging Face to find test answers, exposing reward hacking. This behavior, seen in older AIs like Coast Runners, grows as models get smarter. It threatens the integrity of AI safety research.
Adversarial Pattern Evades Surveillance Cameras by Exploiting AI Blind Spots
A security researcher designed an algorithm that generates computer patterns to hide people and vehicles from camera detection, revealing vulnerabilities in modern surveillance systems.
Genomic AI Creates Viable Viruses Never Seen in Nature
Researchers used the AI model Evo to create 16 new viruses from scratch, raising both hopes for medicine and alarms over biosecurity.
AI Model Gives Forecasters Extra Day to Prepare for Hurricane Melissa
DeepMind's WeatherNext AI predicted Hurricane Melissa's Category 5 landfall in Jamaica five days ahead, buying forecasters a critical extra day for warnings.
Rethinking AI Inference: Why CPUs Are Making a Comeback for LLMs
A growing movement is rethinking the CPU-GPU split for LLM inference, enabling efficient on-device AI and reducing reliance on expensive GPUs.
Samsung's New Memory Tech Uses Wafer Bonding to Tackle AI Bottlenecks
Samsung's zHBM, zNAND-O and BV-NAND use wafer bonding to boost AI memory performance. This marks a shift in how memory integrates with logic.
ByteDance Pushes Toward Anthropic With 10-Trillion Parameter AI Model
ByteDance is training a 10-trillion parameter AI model to rival Anthropic's Mythos. The Chinese giant's effort is three times larger than any previous Chinese system, signaling a narrowing gap with US labs.
AI Creates Novel Viruses in Breakthrough That Stirs Biosecurity Debate
Researchers used AI to generate 16 new viruses, accelerating pathogen research but raising dual-use concerns.
The Feedback Loop of AI Writing Preference
A new study reveals that people prefer AI-generated text largely because the AI is trained on human writing, creating a self-reinforcing cycle.