AI news — March 19, 2026
The 15 top stories PulseAugur surfaced that day, ranked by signal across labs, papers, and developer communities.
-
Startups leverage optical metamaterials to boost AI data center performance
Startups Neurophos and Lumotive are leveraging optical metamaterials, initially explored for "invisibility cloaks," to enhance AI data center infrastructure. Lumotive has developed a new microchip using programmable liquid crystal elements and copper structures that can precisel…
-
80,000 Hours adapts 'Sable scenario' to mirror current AI capabilities
80,000 Hours has adapted the "Sable scenario" from the book "If Anyone Builds It, Everyone Dies" into a video format, making specific changes for simplicity and to better align with current AI capabilities. Their version of Sable reasons in tokens and runs as multiple copies, un…
-
FDE job postings surge, but role's appeal wanes amid AI-generated code issues
The role of Forward Deployed Engineer (FDE) is seeing a surge in job postings, yet it appears to be becoming less desirable among professionals. This trend is partly attributed to the role often resembling a solutions architect position rather than a traditional engineering one.…
-
Replit Agent 4 overhauls design, collaboration, and build workflow
Replit has launched Agent 4, a significant update to its AI-powered development platform. The new version introduces a Design Canvas, replacing the previous Design Mode, which offers an infinite workspace supporting various artifact types and live previews. Collaboration has bee…
-
Canary launches AI QA tool that outperforms GPT-5.4 and Claude Code on code verification
Canary, a new AI-powered QA tool, has launched to automate testing for pull requests by understanding codebases and generating end-to-end tests for user workflows. The tool aims to catch regressions before code merges, addressing a gap in current AI coding assistance. Canary als…
-
AI agent defends consultancy's business model against architect's critique
A building design consultancy owner has developed an AI agent, dubbed 'the talker,' to handle client inquiries and replace the need for junior staff. The agent, built over four months using a duct-taped stack including DeepSeek-R3, aims to improve responsiveness through techniqu…
-
KittenML releases lightweight, CPU-optimized TTS models
KittenML has released version 0.8 of its open-source text-to-speech library, Kitten TTS. This update introduces three new models with parameter counts of 15M, 40M, and 80M, offering sizes from 25MB to 80MB. The library is designed for efficient CPU-based inference without requir…
-
Apple's Mac hardware sees surge in demand for local AI inference
Azeem Azhar has revised his view on Apple's role in AI, recognizing the significant demand for its hardware for local AI inference. Despite Apple's perceived slowness in AI development compared to peers, its Mac devices are experiencing shortages due to their suitability for run…
-
Virtual organ twins allow surgeons to practice complex procedures before operating
The Living Heart Project, launched in 2014, has developed virtual twin technology that creates dynamic digital replicas of patients' hearts from medical scans. This allows surgeons to virtually practice complex procedures, leading to improved outcomes and reduced uncertainty. Bo…
-
NIST reports to Congress on construction safety investigations, including Champlain Towers collapse
The National Institute of Standards and Technology (NIST) has submitted its annual report to Congress detailing progress on construction safety investigations during fiscal year 2025. The report highlights significant advancements in the investigation of the 2021 Champlain Tower…
-
OpenAI uses GPT-4.5 Turbo to monitor internal coding agents for misalignment
OpenAI has developed an internal monitoring system for its coding agents, utilizing GPT-4.5 Turbo to detect and flag potentially misaligned behaviors. This system analyzes agent interactions and chains of thought, alerting to actions that deviate from user intent or violate secu…
-
METR researchers simulate 200-hour AI agents, finding 3-5x workflow uplift
Researchers at METR conducted a tabletop exercise simulating the use of AI agents with a 200-hour time horizon, projecting capabilities expected in about 12-18 months. The exercise aimed to understand emerging workflows and potential productivity gains. Participants found that A…
-
How People ask Claude for personal guidance
Anthropic has released research detailing how users seek personal guidance from their AI assistant, Claude. The study analyzed one million conversations and found that approximately 6% involved users asking for advice on health, career, relationships, and finances. To improve AI…
-
Scott Alexander explores philosophical concepts through fictional characters named John Rawls
This article explores a fictional narrative centered around multiple individuals named John Rawls, creating a humorous and thought-provoking scenario. The story introduces John Rawls the alcoholic, who struggles with declining charity and seeks help from the John Rawls Foundatio…
-
OpenAI acquires Astral to enhance Codex with developer tools
OpenAI has announced its intention to acquire Astral, a company known for its popular open-source Python developer tools like Ruff and uv. This acquisition aims to enhance OpenAI's Codex ecosystem, enabling AI agents to integrate more deeply into the software development lifecyc…