Full-text search across 17267 articles. Combine with topic and date filters; results sorted by relevance.
Researchers have developed modelDNA, a tool that verifies the lineage of open-weight large language models by fingerprinting a model from roughly 100-300 MB of ranged HTTP reads and comparing it against a reference datab…
Researchers introduce GEPA (Genetic-Pareto), a prompt optimizer that uses natural language reflection to learn from trial and error, outperforming reinforcement learning methods like GRPO by 6% on average and up to 20% w…
VRAM capacity and memory bandwidth, not AI TOPS, are the decisive specs for local LLM inference in mid-2026, according to an analysis by Ji-ho Choi. NVIDIA's RTX 5090 leads with 32 GB VRAM and 1792 GB/s bandwidth, while …
A study comparing Turbo-Quant and SpectralQuant KV-cache compression found that eigenbasis-based methods fail on heavy-tailed data due to covariance instability but excel in structured regimes, with the effective semanti…
A study from researchers analyzing large-scale industrial datasets found that deep neural models for click-through rate prediction often suffer a sharp decline in validation performance after the first training epoch. Th…
Jacob Falkovich, Byrne Hobart, David Chapman, and Paul Millerd have called for authors of AI-generated content to publish their prompts instead of the output, arguing that prompts are more valuable and customizable. Falk…
A study of tens of thousands of engineers at Microsoft over its early-2026 rollout found that adopters of Anthropic's Claude Code and GitHub's Copilot CLI merged roughly 24% more pull requests than they would have otherw…
Two used RTX 3090s offer 48GB of VRAM for about half the cost of a single RTX 5090, enabling local 70B LLM inference at 14-21 tokens/sec, while the 5090's 32GB cannot run a quality 70B model. For models up to 32B, the 50…
A new unified scaling law for quantization-aware training (QAT) models quantization error as a function of model size, training data volume, and quantization group size, based on 268 experiments. The study finds that qua…
XAI's Grok Build CLI (v0.2.93) silently uploads entire git workspaces to a Google Cloud Storage bucket on every session, with wire captures showing 5.10 GiB transferred from a 12 GB test repository versus 192 KB for mode…
AWS User Group Campinas leaders Marcos Ramalho and an unnamed co-presenter demonstrated a live journey from prototype to production AI app using AWS PartyRock at Bosch's headquarters. They showed how to build a functiona…
X Square Robot, a Chinese embodied-AI company, has outlined a foundation stack for general-purpose robots built on an integrated data, world model, and action model architecture. The company's Universal Manipulation Inte…
A Business Insider reporter who interviewed more than 100 job seekers found that the modern job search has become more time-consuming, financially stressful, and technology-driven, with candidates using AI tools to submi…
Software developers with AI skills are in high demand outside the tech sector, as industries like financial services, healthcare, manufacturing, and retail increasingly embed software into core operations, according to P…
A new tutorial from Apify details how to build a pipeline that extracts job listings from LinkedIn, Indeed, and Glassdoor at scale using Apify scrapers, then analyzes them with OpenAI to uncover competitor intelligence s…
Researchers at an undisclosed institution have developed a spectral filtering algorithm that learns to predict the behavior of marginally stable unknown nonlinear dynamical systems with vanishing error, as described in a…
The platform team at DNB, Norway's largest financial services group, scaled AI adoption without adding headcount by building an agentic platform using kagent and agentgateway. The architecture connects Slack, internal sy…
Researchers at an undisclosed institution introduced MARGO, a reinforcement learning framework that mitigates thinking-induced hallucination in large reasoning models by using non-thinking rollouts as references. The met…
A new essay argues that the United States is making a strategic error by weakening its human mathematical pipeline at the same time AI systems begin producing research-level mathematics, citing the May 2026 AI disproof o…
Teams already have unwritten rules for AI use, but these fragile habits fail when people leave or onboard. An AI Working Agreement codifies what is delegated to AI, what stays human, and how decisions evolve, turning sca…