Claude 3.5 Sonnet outperformed GPT-4o, Gemini 1.5 Pro, and DeepSeek-V3 in Spring Boot configuration and code generation tests on a legacy Java codebase, correctly identifying transaction boundaries an…
VentureBeat has hired its first analyst, Strechay, to lead research on enterprise AI infrastructure, cloud, and platform engineering, with a focus on providing empirical data to enterprise builders an…
Kimi K3, an AI model from Moonshot AI, introduces Attention Residuals, replacing standard uniform residual connections in deep transformers with attention-weighted layer outputs, allowing the model to…
Anthropic's Claude AI model achieved a 93% success rate in designing functional de novo protein binders for 14 of 15 targets, with wet-lab validation by Adaptyv Bio and Twist Bioscience, according to …
A new analysis identifies the visual and technical fingerprints of 'vibe coded' websites, including default palettes of charcoal and neon, Inter or Geist fonts, 12px or 16px rounded corners, and Shadc…
An open-source MCP server that runs on Android and exposes tap, type, scroll, and screenshot actions to AI agents has been released on GitHub, with a FOSS build heading to F-Droid. The app, which regi…
Anders Hejlsberg unveiled a Go-based rewrite of the TypeScript compiler that processes the VS Code codebase (roughly 1.5M lines) in under a second, compared to 12-15 seconds for the old implementation…
Cerebras Systems' CS-4 rack achieves 85% model flops utilization (MFU) on a 400B parameter training run across 16 systems, compared to 30-40% for traditional GPU clusters, thanks to the WSE-3's on-waf…
Relay, a startup that built AI agents capable of operating web interfaces, has been acquired by Google, according to a blog post. The team's technology, which handles dynamic DOMs, shadow roots, and o…
Zed Editor's native AI features, built into its Rust-based text engine, outperform Cursor and GitHub Copilot in latency benchmarks, with single-line completions at 12ms locally and 180ms via cloud, ve…
The volume of unsolicited book queries has surged from about 200 per month to 200 per week since ChatGPT went mainstream, with most submissions now AI-generated or heavily AI-assisted, according to in…
Developers seeking practical AI coding help are finding more value in tool-specific Discord servers and specialized forums than in Reddit or Hacker News, according to a developer's account. The articl…
A JAMA viewpoint argues that autonomous AI will outperform every doctor-AI team, citing automation bias and human latency as the weak link once model performance exceeds a threshold. The authors conce…
Parallel, Exa, and Firecrawl lead the Artificial Analysis Search Index for search APIs, with Parallel winning on quality, Exa on speed and cost, and Firecrawl on free-tier accessibility, according to …
Most companies will miss the AI-native shift because they bolt LLMs onto legacy systems, resulting in 'smart dumb processes,' according to an analysis of three enterprise case studies. The piece argue…
Voice cloning technology has made grandparent scams more convincing, with open-source models like Bark allowing anyone to clone a voice in under two minutes, according to a first-person account. The a…
A developer built Relay, a remote control tool for AI agents running on a home server, enabling live terminal access from any device with session persistence, file transfer, and token quota tracking. …
Axiom Math's AI system, AxiomProver, has verified the 246 theorem on prime gaps, building on Yitang Zhang's 2013 proof and James Maynard's refinement to 600, and creating a reusable library of results…
AI News Digest for Aug 19 covers 30 stories, including Galbot preparing a new humanoid for August robot showcases, Leopold Aschenbrenner losing $30 billion by betting too heavily, and State Farm lawye…
Galbot is preparing to debut a new humanoid robot at the World Robot Conference 2026 (WRC26) in August, aiming to demonstrate real-time capabilities in competitive tasks. The company's reveal will tes…