Research & Benchmarks - Page 19 of 35
Academic AI research, performance benchmarks, scientific breakthroughs, and peer-reviewed studies advancing artificial intelligence frontiers.
698 articles
View complete article list
Fine-tuning RAG embeddings may drop retrieval accuracy 40%, study finds
Why does this matter? Companies deploying retrieval‑augmented generation (RAG) often chase tighter precision by tweaking the underlying embedding layers, assuming tighter vectors will feed cleaner results to downstream agents.
AI pipelines show silent failures from orchestration drift, detected weeks later
An AI pipeline can look flawless in testing. Under real-world load, something subtle shifts. No crash. No alert. Just a quiet divergence, the sequence of retrieval, inference, tool use, and downstream action begins to drift.
OSWorld Benchmark Evaluates LLMs on Real Computer Use, Unlike Text‑Only Tests
Most AI agent tests live in a walled garden of text and APIs. Not OSWorld. Princeton researchers built it to throw models into a real desktop environment—a full computer, with no shortcuts allowed. The result? A stark 60-point performance gap.
PageIndex Retrieves via Reasoning Using OpenAI gpt-5.4 Model
We've been building better search the wrong way. For years, retrieval meant vectors. You'd smash text into dense embeddings, throw them into a specialized database, and hope the nearest neighbor was the right answer.
Discord Users Access Anthropic's Mythos AI Tool Without Authorization
Anthropic called its Mythos AI too dangerous to release. That was the official story. On Discord, a different narrative emerged. A handful of users, sifting through a leaked dataset from the training startup Mercor, made a simple guess.
Google DeepMind's Vision Banana Outperforms SAM 3 and Depth Anything V3
Imagine a single model that looks at an image and does it all, segments objects, measures depth, reads surface angles, and does it better than the specialists built for each. That’s the leap Google DeepMind just pulled off with Vision Banana.
DeepMind spinoff’s AI‑designed drugs enter human trials after AlphaFold 3
The protein-folding models were always destined for the real world. That was the whole point. Now a spinoff from DeepMind is pushing drugs designed by its Nobel-winning AI into human trials.
COALA paper defines agent memory types: procedural rules and semantic facts
The AI industry is obsessed with memory. It's treated as a mystical new organ, demanding custom architecture and bespoke brain-lobes. Then a formal academic document, the COALA paper, dropped a deflatingly simple taxonomy.
Google DeepMind's Decoupled DiLoCo hits 88% goodput despite hardware failures
The dream of training AI at planetary scale has always collided with a brutal physics problem: moving data between continents is slow, expensive, and unreliable.
Agent observability powers production evaluation through trace analysis
Your agent performs flawlessly in staging. Then reality hits. Users do the unexpected, they ask offbeat questions, skip steps, or trigger edge cases no one considered, and suddenly your model’s output unravels.
Xiaomi launches MiMo‑V2.5‑Pro and V2.5, matching benchmarks at lower token cost
Xiaomi just dropped two new AI models. They hit the same performance marks as the industry's top benchmarks. But they cost less to run. The flagship, MiMo-V2.5-Pro, now leads Xiaomi's own MiMo Coding Bench.
Designing Production-Grade CAMEL Multi-Agent Systems: Start with Docs and GitHub
Everyone wants multi-agent systems to just work. They don't. The trick isn't building a new world from scratch. It's spelunking through the mess someone else already made. For CAMEL, that means the docs and the GitHub issues are your new bible.
Multi-agent AI systems incur higher token costs than single agents in practice
Multi-agent AI is a brute-force lie. We see a neat line of specialized chatbots, each handing off a problem, and mistake the pageantry for efficiency. It's just throwing more computer power at the wall.
Reinforcement learning trains AI like OpenAI's o1 to admit uncertainty
OpenAI's new o1 model will answer any question put to it. It will do so with terrifying, unblinking certainty. This is by design. The reinforcement learning that built it pays for correct answers, full stop.
LangSmith adds reusable LLM-as-judge and rule-based code evaluator templates
Every production AI application is a live experiment. Until now, keeping that experiment under control meant duplicating evaluators across projects, rewriting prompts, and hoping nothing slipped through. LangSmith changes that.
AI made up over a third of new sites by 2025; Pope warning flagged as AI
AI-generated text now accounts for more than a third of new websites. That figure comes from researchers at Stanford University, Imperial College London, and the Internet Archive, who published their findings this month.
Sergey Brin pushes DeepMind to match Claude, unveils agent skills catalog
Sergey Brin is personally on the bench at Google. That fact alone signals a crisis. The billionaire co-founder has one blunt mission for DeepMind: catch Anthropic's Claude. In naming that target, Google has quietly admitted the new bar. Claude is it.
Fortnite adds AI‑powered NPCs for unscripted player conversations
Fortnite’s computer-controlled characters have always sounded like, well, computers. That’s about to change.
TabPFN hits 98.8% accuracy in 0.47 s, beating Random Forest and CatBoost
TabPFN is a blunt instrument. It clocks 98.8% accuracy in 0.47 seconds. That number isn't a minor improvement over Random Forest and CatBoost, it's a different league. The trick is that it does no training. It just sets up.
NVIDIA PhysicsNeMo Tutorial Maps k(x,y) to u(x,y) for Darcy Flow
Mapping a heterogeneous permeability field to a pressure distribution , that’s the core of Darcy flow, a fundamental problem in subsurface modeling and reservoir engineering. Traditional solvers are accurate but slow.
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.