Category
60 articles in this category
In this tutorial , we explore the Open-SWE-Traces dataset as a practical resource for studying and preparing agentic software-engineering trajectories for fine-tuning.
A new Cursor study reports that newer coding agents often retrieve known fixes instead of deriving them, inflating popular benchmark scores.
MRAgent framework reduces token consumption and runtime costs for long-horizon reasoning tasks in AI agents.
OpenAI limits GPT-5.6 release to a small group of trusted partners at US government behest.
Incomplete data hampers autonomous security agents; experts stress verifying endpoint coverage before deployment.
Anthropic's Mythos-class models stay offline two weeks after Trump administration ultimatum.
It’s been hot in London this week.
Anthropic alleges Alibaba launched largest attack to clone AI model Claude, violating terms of service and access restrictions.
The drive to discover the next big thing in AI has funded some pretty ambitious projects — but one company is taking it as a chance to rebuild computing architecture from the ground up.
Most major AI chatbots lean left on political questions, according to a Washington Post investigation.
Insurers use diffusion models to generate plausible weather events for risk assessments, but face warnings about hallucinations.
The US government and Colossal Biosciences partner to sequence genomes of all species on the Endangered Species list.
In the mid-noughties, when music by the Killers and Franz Ferdinand blared out of every pub and nightclub I passed, I spent my days and nights struggling through a Ph.D.
Most end-to-end OCR models slow down as output grows.
New data contradicts the notion that AI is replacing engineering jobs, instead showing they're resilient and in high demand.
AI researchers Jonas Adler and Alexander Pritzel leave Google for Anthropic, continuing a trend of top talent departing for rival companies.
Rep. Anna Paulina Luna says staff used AI for 'spellcheck' in amendment summary, but denies AI wrote bill text.
Alibaba's Qwen team releases Qwen-AgentWorld, two models trained to predict environment returns, boosting agent performance in seven domains.
In this tutorial , we build OpenHarness from scratch to better understand how a practical agent harness works.
Xiaomi's HarnessX framework autonomously improves AI software scaffolding, yielding significant performance gains across domains.
Stanford researchers develop agentic AI 'scientists' to streamline drug discovery, reducing inefficiency and high failure rates.
Amazon's AGI autonomy lab develops framework for trustworthy AI agents, focusing on consistency, robustness, and safety.
AI helps read 2,000-year-old charred scroll, revealing stoic philosophy on ethics, art, and human behavior.
Summary RFIC design is a complex “ dark art ” that limits progress in wireless technologies like 5G, autonomous vehicles, and satellite communications.
AI is booming.
The writer who coined the word ‘enshittification’ tells us why AI will never deliver what it promises – and why it still appeals so much to those in power A “centaur”, in automation theory, is a person assisted by a mac…
Autoregressive large language models generate text one token at a time.
🚀 First open far-field ASR benchmark: community-driven evaluation across 14 simulated rooms, validated against real-world measurements: https://huggingface.co/spaces/treble-technologies/ffasr 📉 The gap is real and it is…
The White House cuts deadline for government agencies to adopt quantum-resistant encryption to protect against quantum computer attacks.
TL;DR — Building an agent is mostly plumbing: tools, state, guardrails, scaling from one agent to many.
Cory Doctorow's new book targets AI hype and explores its implications on society
Imagine sitting down at your desk and logging in for a performance review, with an AI system analyzing the conversation.
Enterprises struggle to scale AI workloads due to fragile data delivery infrastructure.
OpenAI teams up with Trail of Bits to help open-source maintainers secure their projects and improve cybersecurity.
Researchers pinpoint origin of long-period radio transients, a mysterious phenomenon of strong radio signals arriving periodically from space.
On Friday, Claude Code creator Boris Cherny made an appearance at Meta’s @Scale conference and, surprisingly, the first question from the audience was about loops.
Artificial intelligence is the transformative, strategic technology of the early 21st century.
OpenAI announces cybersecurity-focused initiatives to combat AI hacking capabilities and partners with Trail of Bits to launch Patch the Planet.
Organizations need to capture and reuse knowledge gained from daily operations to improve AI-driven decisions.
Self-Harness lets AI agents systematically improve their operating rules, boosting performance up to 60% without relying on human engineers or stronger external models.
MoonMath AI team has released a bf16 forward attention kernel for AMD’s MI300X GPU.
As AI inference workloads evolve, GPU availability is no longer the primary bottleneck; instead, context management has become a major challenge.
Large language models are stateless by default.
Getting prompts right is still the hardest part of shipping reliable LLM applications.
Signal President Meredith Whittaker cautions against treating AI chatbots like friends or sentient beings, citing serious privacy implications.
The Atlantic's Alex Reisner uncovers four datasets of music used to train AI models, making them searchable for the public.
Thought experiment about US AI ascendancy sparks concerns about Europe's preparedness
NVIDIA Research has released SpatialClaw, a training-free framework for spatial reasoning.
While recent breakthroughs in AI reasoning have largely been driven by massive scale, pouring in billions of parameters to cross complex cognitive thresholds— VibeThinker-3B is charting a completely different path.
Nobel Prize winner John Jumper leaves Google Deepmind for Anthropic after nearly nine years.
Enterprise AI agents often stall in production, requiring human oversight, but hypernetworks may offer a solution by generating task-specific models on demand.
10% of people worldwide now use AI chatbots for news weekly, up from 7% last year, but trust remains low.
Top AI model solves just 3% of realistic knowledge work tasks, highlighting significant challenges in handling complex projects.
Subquadratic claims to have solved a mathematical bottleneck in large language models, while BCI trials surge and a man with ALS uses a brain implant.
Most AI memory remembers the user.
Deep research agents increasingly combine private local documents with external tools like web retrieval, creating a privacy risk: an agent's external queries may leak sensitive information.
KV cache compression methods, including TurboQuant, OSCAR, and EpiCache, aim to reduce memory usage in long-context large language models.
Underneath an Apennine massif, below the Jinping Mountains of Sichuan, and at the bottom of a South Dakota mine, there is a cosmic hunt afoot.
Noam Shazeer, co-author of 'Attention Is All You Need' and former Google Gemini co-lead, joins OpenAI after a two-year return to Google.
OpenAI's LifeSciBench evaluates AI models on real life-science research tasks with expert-written rubrics.