AI News Archive - Browse Page 22 of 182
Browse AI news articles covering LLMs, tools, research, and industry trends
DFlash drafts whole token blocks, achieving 15× throughput on NVIDIA Blackwell
Token-by-token generation is the bottleneck that has kept large language models tethered to a serial fate. DFlash breaks that chain.
RIFT-Bench Introduces Graph-Driven Dynamic Red-Teaming for Agentic AI
Testing AI agents for security holes is a manual, brittle mess. Each new framework demands a custom audit; crafted attacks are often obsolete before...
Survey of AI Agents: Descartes, Sci‑Fi Roots, and Current Architectures
For centuries, philosophers and novelists have argued over what makes an agent. The fight is no longer academic.
Mistral OCR 4 Delivers Citation‑Ready Structured Output for RAG and Search
Data is the lifeblood of modern AI, but raw document text is a hemorrhage of noise.
Krea 2 Raw/Turbo generate AI images in 2 s; Nano Banana Pro 17.7 s proprietary
Two seconds. That’s all it takes for Krea 2 Raw and Turbo to generate an enterprise-grade image.
Correlated errors cut panel accuracy 8‑22 points; top judge matches panel
We keep stacking AI judges into panels, hoping a crowd of models will be wise. According to new research from Apple, it isn't.
Metric-Dependent Annotation Saturation for Learning from Label Distributions
We treat annotator disagreement like static to filter out. Maybe we're filtering out the point.
Agentic observability unites telemetry to cut incident investigation time
Outage investigations follow a predictable, expensive script. Someone shouts, everyone scrambles for logs, and the clock ticks on revenue and...
DFlash speculative decoding boosts NVIDIA Blackwell inference up to 15×
NVIDIA Blackwell delivers 15 petaflops of dense NVFP4 compute, a staggering amount of raw power.
NVIDIA architectures boost AI per‑watt efficiency with full‑stack optimizations
NVIDIA built its AI empire on raw speed. Now, the glaring constraint is the power bill.
NVIDIA BioNeMo Toolkit Enables AI Scientist to Align, Fold, and Dock Molecules
Lab work is slow. The AI scientist is not. It runs without sleep or salary, folding proteins and aligning sequences in a silent digital loop.
OpenAI's GPT-5.5-Cyber Beats Anthropic Mythos, Starts Patching Initiative
OpenAI's GPT-5.5-Cyber just beat Anthropic's Mythos on key cybersecurity benchmarks. That’s the flashy result. Look past it.
Pull Gemma4:e4b with Ollama to Build a Local AI Coding Agent (v9.6)
The 9.6 GB download is a promise. A 128K context window, a 4-bit quantized model, and the raw power of Gemma 4 sitting right on your NVIDIA RTX 2000...
NVIDIA OpenShell Secures Agentic AI in Telco Autonomous Networks
The telecom industry dreams of networks that run themselves. It’s a powerful fantasy: AI predicting a cell tower’s failure before it happens,...
GLM-5.2 API guide emphasizes tool‑based lookups, not guesswork
Guesswork is a luxury no serious analyst can afford. Numbers demand precision, not approximation.
AI research increasingly relies on recursive loops, a staple of CS basics
Every comp-sci freshman learns recursion: a function calls itself, cracks a problem, and stops. That's the textbook version.
xAI adds /goal to Grok Build for autonomous multi-step coding with verification
Code that writes itself is a trick. Code that writes itself, then double-checks its work, then only declares success when it’s actually right, that...
Alibaba AI video model climbs to #2 as Sora withdrawal warns firms
Alibaba’s AI video model, dubbed HappyHorse, has surged to second place in the global Arena rankings, nudging past Google’s Veo 3.1 and capitalizing...
Sakana AI launches Sakana Fugu; Fugu Ultra leads coding, reasoning and tests
The numbers do the talking: Fugu Ultra crushes four coding benchmarks, tops CharXiv Reasoning, and aces Humanity’s Last Exam.
Anthropic's government feud: three warning signs and a superficial response
Regulators shut down Anthropic's Fable model days ago. The company's official statement still lacks a real safety plan.