> AI_ENGINEER.NEWS

VideoLAN launches dav2d, a fast, open-source AV2 decoder inspired by dav1d, to accelerate AV2 adoption across platforms.

New Bonsai Image 4B models enable high-quality, memory-efficient local image generation on iPhones and devices, transforming deployment and privacy.

Overusing AI tools erodes focus and productivity, leading to unmaintainable projects and superficial engagement; reducing use fosters meaningful work.

ChatGPT for Google Sheets is vulnerable to prompt injection, enabling data theft and phishing attacks even with security settings enabled.

Odysseus v1.0 is a self-hosted AI workspace serving local models, tools, and integrations — privacy-focused, versatile, fun, with extensive features.

AI accelerates prototyping and coding, boosting productivity 4x, enabling faster idea testing, but requires deliberate skill maintenance and careful planning.

Some believe in AI replacing humans as the next evolutionary step, challenging human-centered ethics and calling for a new humanism.

AI-generated influencers fake Black women to sell cheap Shein products, exploiting race and emotion for dropshipping scams on TikTok.

Use of LLMs in software development is increasing productivity but reducing system throughput and quality, ultimately destroying organizational value.

Remove LLM-generated commits to prevent potential issues and associated risks in the codebase.

Claude reduces coding but boosts testing and understanding, transforming development into more logical, exploratory work.

Komi-learn enables AI agents to self-improve and remember sessions automatically, integrating with Claude and Codex.

Humans often misuse intelligence, leading to unpredictable, hallucinative, and misaligned thinking similar to stochastic parrots.

Connecticut mandates AI transparency in hiring, layoffs, bias testing, social media use, and expands AI workforce education.

Julia’s GPU-accelerated ODE solvers are 20x-100x faster than Jax and PyTorch due to architectural differences in GPU implementation.

AI systems based on unlawful web scraping violate human rights, threaten privacy, and should be prohibited, says Amnesty International.

Netflix engineer creates open-source app, Headroom, reducing AI token costs by up to 90%, saving money and improving model efficiency.

AI and LLM hype mirror the dotcom bubble, with lofty promises, flawed tech, but the core tech remains influential and evolving.

Open-weight AI models are increasingly accessible, easily stripped of safety guardrails, posing risks for misuse, harm, and malicious activities.

Ouijit is an open-source, customizable task and terminal manager for coding agents with live status, sandboxing, and CLI integration.

Tech elites pursue transhuman AI to conquer cosmos, risking societal neglect, inequality, and unchecked power in a secular, sci-fi-inspired religious quest.

Memory-as-Action (MemAct) optimizes working memory for long tasks, matching larger models' accuracy while significantly reducing context length.

AI coding agents are expanding terminal roles, requiring better UI to manage multi-task, state, and provenance for supervision.

UChicago developed Quicksilver, a browser extension that detects AI-generated music, promoting transparency and protecting human artists.

U.S.-China AI battle shifts from development to control, as China distills U.S. models and risks global dependence, security, and leadership.

A CPU-only, offline OCR tool that extracts text from screenshots, images, PDFs, and web pages with a single command.

AI models struggle to update beliefs with new evidence, risking unreliable reasoning in science and critical tasks.

UC Berkeley Law bans AI for coursework and exams in 2026 to prioritize critical thinking and legal skills over AI reliance.

Chiefs emphasize respectful AI use, focusing on human review, brevity, and empathy, rather than just policies, to foster team trust.

Stable Audio 3 enables fast, batch sample generation for drums and pitched instruments on GPU, with quality gating, retries, and customizable prompts.

New pricing for Copilot may be a joke; users face network blocks and need to file tickets to resolve access issues.

Local AI names macOS screenshots automatically, securely, and privately using bundled models without external upload.

AI-generated code using Codex bypasses security constraints, highlighting potential risks in AI-driven code development.

Run and manage multiple AI coding agents locally, chain tasks, rotate accounts, schedule routines, and sync configs—fully open-source.

HarnessKit centralizes management of AI agents, extensions, configs, and security across multiple platforms and projects efficiently.

AI tool ESMFold2 predicts 1.1 billion protein structures, surpasses AlphaFold, and aids in new biology discoveries.

A free cloud tool to manage AI agents across multiple hosts, assign tasks, chat, track costs, and centralize workflows.

PewDiePie's open-source Odysseus is a self-hosted AI workspace with chat, tools, research, and private data tracking, all locally run.

Max creates algorithms to randomize decisions, seeking freedom from routine but risking aimlessness and avoidance of responsibility.

SIA is an AI system that autonomously updates its model weights and harness during inference to improve performance across tasks.

Open-source HeartMuLa models enable multilingual music generation, transcription, and data alignment, matching top commercial fidelity.

OMP is a native, extensible coding agent with embedded IDE, tools, real debugging, memory, web access, and subagents, all on local machines.

AI agents will use DNS for discovery, enabling scalable, secure, and decentralized communication without hardcoded configs.

Claude and Codex now support real-time Git conversations, enhancing AI coding interactions.

AI-driven UX testing identifies key web issues quickly, helping teams improve user experience and business metrics.