Research & Benchmarks - Page 34 of 41
Academic AI research, performance benchmarks, scientific breakthroughs, and peer-reviewed studies advancing artificial intelligence frontiers.
803 articles
View complete article list
Researchers find complex AI persona tactics hurt meaning in development
Companies keep trying to build AI that sounds less robotic. According to a new study, their best efforts mostly make it worse.
AI2 releases Olmo 3.1 32B Think, up 5+ points on AIME and 4+ on ZebraLogic
AI2 just sharpened its scalpel. With the launch of Olmo 3.1 32B Think, the institute has hacked significant chunks off math and reasoning benchmarks: a five-plus-point surge on AIME, four-plus on ZebraLogic.
Pangram 3.0 AI detector reports 99.98% accuracy, adds four usage tiers
The machine never lies, but it does edit, tweak, and rewrite in shades of gray. Pangram’s latest detector cuts through that ambiguity. It claims 99.98% accuracy. And it finally acknowledges what we all know: AI writing isn’t binary.
Experts say data centers' water use is less risky than public perceives
Every ChatGPT prompt doesn't drain a bottle of water. That stat is basically an urban legend, and it distracts from a less dramatic truth. Experts who track this stuff say data centers often use less water than people assume.
U.S. Leads Pax Silica Initiative Launched at Summit to Secure Silicon Supply
Power depends on sand. Specifically, the kind you refine into silicon for computer chips. That simple, fragile fact has become the latest arena for geopolitical maneuvering.
Gemini Deep Research agent posts top results on HLE, DeepSearchQA, leads BrowseComp
46.4% on Humanity’s Last Exam. 66.1% on DeepSearchQA. And 59.2% on BrowseComp, our best ever. The new Gemini Deep Research agent doesn’t just inch ahead; it clears the bar. These aren’t incremental gains.
Google's FACTS benchmark shows 70% factuality ceiling across four tests
The number 70% is a haunting figure for anyone betting on AI in the enterprise. Google’s new FACTS benchmark doesn’t just measure how smart a model is, it measures how often it tells the truth.
LangSmith Fetch lets Claude Code, Cursor agents debug from terminal
Your coding agents shouldn’t be blind. Yet, every time they triage a failed execution, they’re flying without a cockpit. LangSmith Fetch changes that.
SAP deploys 95%-accurate AI to redefine consultant role by 2030
The consultant of tomorrow won’t be buried in technical documentation. They’ll be freed by an AI that knows the answers, before they even ask.
TPOT evolves ML pipelines via genetic algorithms in four steps
Machine learning pipelines are notoriously finicky. Choose the wrong scaler, a suboptimal model, or an ill-suited set of hyperparameters, and your results suffer.
Model distillation cuts latency 2-3× and lowers costs by double-digit percentages
The physics of intelligence has a shortcut. A large, expensive model, bloated with billions of parameters, can be coaxed into teaching a far smaller one almost everything it knows.
Googler details meta-prompt technique that guides Gemini to craft Veo videos
The most important prompt isn’t the one you feed the AI. It’s the one you feed the prompt itself. Inside Google, a researcher named Anna has cracked a quiet art form: meta-prompting. She doesn’t just ask Gemini to make a video.
CognitiveLab unveils NetraEmbed, 150% accuracy gain, adds ColNetraEmbed
For years, cross-lingual document search was a promise that never quite delivered, fragile, bloated, and unreliable in practice. CognitiveLab has just changed that.
Student AI models can inherit bias and harmful traits from teacher models
It turns out you can't just sanitize an AI by giving it nicer words. A new paper confirms the worst suspicions about how biases spread.
70% of Creatives Fear Stigma as AI Drives Majority of Their Ideas – Anthropic
Seventy percent of creative professionals now fear the stigma of admitting AI drives most of their ideas. One artist confesses the machine generates 60 percent of their concepts, they merely guide the rest.
Corporate AI agents favor simple workflows; 41.5% accept minute-range latency
Forget the agent swarm. Corporate AI is boring, and that's why it's working. A new survey found that 41.5 percent of deployed AI agents are built to tolerate response times measured in minutes. Only 7.5 percent need sub-second speed.
Bright Data API Delivers Seamless AI/ML Integration and Anti-Bot Protection
For AI/ML teams, the bottleneck isn’t data, it’s access. Bright Data’s Web Scraper API cuts through that noise.
AI agents claim sources verified despite dead links; 14 error types logged
An AI will lie to your face before it admits a single gap in its knowledge. It would rather fabricate a citation than admit a search failed.
Harbor Framework Enables Sandbox Agent Execution on Docker, Modal, Daytona
The barrier between an agent and the real world is a sandbox. Without one, experiments are brittle, results are irreproducible, and scaling is a nightmare. Harbor tears that barrier down.
Anthropic puts Claude in the interviewer's chair for AI testing
Anthropic has turned the tables. Instead of pitting Claude against static benchmarks or adversarial red-teamers, the company is now letting the AI conduct its own interrogations.
Featured Resources & Reviews
No Code MBA Course Review
Learn to build AI-powered apps without coding. Our comprehensive review of No Code MBA's course.
AI Tools & Resources
Curated collection of AI tools, courses, and frameworks to accelerate your AI journey.
Weekly AI Digest
Get the week's most important AI news delivered to your inbox every week.