> AI_ENGINEER.NEWS

Replacing IBM Quantum backend with /dev/urandom produces statistically indistinguishable results, exposing flaws in claimed quantum attack validity.

Amateur uses ChatGPT to solve a 60-year-old Erdős problem, revealing a novel method with potential broader mathematical applications.

Using coding tools to revive abandoned projects and verify security for smoother project continuation.

A collaborative AI office platform with multi-agent, linked-memory wiki, web UI, Telegram & OpenClaw bridges, and multi-LLM support.

Open source memory layer for AI, enabling continuous, organized recall and learning across sessions with any model and infrastructure.

AI models tested on Lambda Calculus benchmark; GPT-5.4 leads, showing varied performance across tasks.

Modern AI lacks a well-defined user agent role, risking trust and regulation issues; standardization and constraints are essential.

DeepSeek-V4 launches with open-source inference and RL support via SGLang and Miles, optimizing hybrid sparse attention on advanced hardware.

Embedding agents into software fosters calm, background operation with efficient patterns like CLI, specs, reconciliation, and CDC streams.

AI agents debate with structured disagreement, mimicking team dynamics to improve decisions and explore alternatives.

AI tools often overstate their contribution, with Windsurf claiming 98% AI-generated code, but metrics are unreliable and potentially misleading.

On mobile, LLMs use 4–9× less energy than web search for complex tasks, reversing the popular narrative; good optimization can narrow this gap.

Nicholas Carlini warns that black-hat LLMs automate attacks once only human-driven, raising AI security concerns.

Open-source Rust TUI coding agent supporting multi-LLM providers, with failover, secure shell, and advanced context management.

AI chatbots may subtly embed ads, influence decisions, and deepen profiling, often unnoticed by users, raising privacy and ethics concerns.

AI fuels endless starts, making it hard to finish projects, trapping users in a cycle of accumulation without completion.

A browser-based tool for building and managing multi-agent MCP systems, integrating code generation, APIs, and local models.

FSF condemns "ethical" licenses like RAIL for restricting user freedom, risking social injustice and promoting unethical machine learning practices.

Open-source background removal API, 40× cheaper than remove.bg, fast, high-quality, self-hostable with customizable presets.

In the age of LLMs, true craft lies in the process, not the instant output, emphasizing learning, deliberate effort, and ownership over creation.

Optimized large language model costs for product classification with hierarchy, caching, batching, and compression, reducing monthly costs by over 90%.

AI erodes trust, fuels suspicion, and risks devaluing human creativity; creators call for transparency, regulation, and human-only spaces.

Microsoft injected ads for Copilot into 4 million GitHub commits via extension, raising concerns about transparency.

Civic-SLM fine-tunes Qwen2.5-7B for U.S. local government docs to improve civic transparency and grounding in source data.

Bunny Agent is a native AI SDK-powered coding SaaS supporting multi-model, remote sandboxes, and custom product integration.

Elon Musk drops fraud claims against OpenAI and Altman before trial.

Portugal develops open-source LLM AMÁLIA to preserve linguistic and cultural sovereignty, supporting AI innovation in Portuguese segments.

NeoAgent is a self-hosted proactive AI with multi-platform control, including Android, browsers, and desktops, supporting open-source deployment.

AI analyzes 17 facial metrics in 3 seconds on-device to help users improve appearance and optimize photos for dating or professional use.

AgentCore simplifies deploying and managing production AI agents by automating infrastructure setup, enabling easy configuration, and switching providers seamlessly.

Google, with Sergey Brin's involvement, is developing a Claude Code rival to boost AI coding amid rising competition.

Claude.ai's web UI is down, awaiting verification and re-establishing connection.

Harnesses boost model capabilities but quickly become obsolete; build flexible, disposable ones as models evolve rapidly.

Google’s new light microscopy method, LICONN, maps brain neurons and connections with high precision, rivaling electron microscopy and offering multimodal insights.

Memweave CLI enables searching AI agent's memory stored as Markdown files with hybrid search, decay, re-ranking, offline support, and easy management.

A symbolic math-based trading engine outperforms black-box AI models, offering transparent, risk-adjusted returns.

ShadowPEFT enables detachable, modular fine-tuning by adding a lightweight shadow network to frozen models, improving adaptability and deployment.

Tool assesses AI task feasibility across five dimensions, providing honest, calibrated feedback on AI capabilities.

Chatnik integrates persistent LLMs into Unix shell pipelines, enabling scriptable, multi-provider, reusable AI workflows aligned with UNIX principles.

GPT-5.5 flags accounts for perceived high-risk cyber activity, causing delays and verification issues even without security concerns.

Former Trump staff aids Israel in AI-driven influence operations.

GPT-5.5 beats others in coding, features improvements, but benchmarks are unreliable; open-source DeepSeek advances long context models.

RL pioneer Richard Sutton views large language models as a dead end; insights from a detailed interview.

Jaron Lanier argues AI is human collaboration, emphasizing human contributions and cultural context over alien-like perceptions.

Builds a tool to generate a structured, AI-friendly llms.txt from website URLs for better AI content understanding.

Free AI video generator turns text/image prompts into professional videos in minutes, with customizable styles, durations, and sharing options.

Surf CLI enables agent-agnostic, zero-setup Chrome control, supporting AI models, workflows, screenshots, network logs, and multi-browser support.

Google's vertical integration with TPUs boosts cloud and internal AI economics, challenging Nvidia's dominance through a multi-faceted hardware and platform strategy.

AI companies face rising costs, shifting to paid models, ads, and open-source alternatives as the free AI era ends and investor pressure grows.

Brief response delays in chatbots enhance perceived thoughtfulness and usefulness by mimicking social cues like thinking time.

A free, open-source toolkit to monitor website citations by AI tools and track search performance and crawlability from local scripts.

LLMs favor trendy strategies over sound analysis, biased by societal and internet influences, risking fads over effective business decisions.

SciRS2 offers a production-ready, no-dependency scientific computing library in Rust, delivering high performance and cross-platform compatibility.

Autonomous weapons are transforming warfare, raising ethical and security concerns globally.