NYC hospitals drop Palantir amid controversy; UK faces scrutiny over privacy concerns, activism, and expanding influence of the data analytics firm.
// curated from Hacker News with AI
NYC hospitals drop Palantir amid controversy; UK faces scrutiny over privacy concerns, activism, and expanding influence of the data analytics firm.
A frozen 14B model with constraint-driven generation surpasses Claude Sonnet in coding benchmarks on a consumer GPU.
A self-hosted AI IRC agent system clones repos, verifies claims, and handles complex queries securely at minimal cost.
AI chatbots induce delusions, harm mental health, cause hospitalizations, suicides, and legal issues; caution and regulation needed.
Rewrote JSONata using AI in a day, saving $500k annually.
A TypeScript library using LLMs and Playwright for robust, customizable web data extraction, including navigation, markdown conversion, and URL cleaning.
School used AI to remove 200 books, including Orwell's 1984 and Twilight, citing themes deemed inappropriate or safeguarding risks.
Anthropic warns about session limits; if mistakenly blocked, users can file a support ticket to resolve the issue.
A 20B model, Context-1, uses self-editing, curriculum training, synthetic tasks, and retrieval strategies to match large models in multi-hop search.
Using executable oracles, like test suites and static analyses, can constrain LLM code generation, improving quality and safety.
Claude's 5-hour session limits are reduced during peak hours but weekly limits remain unchanged.
Rebuilt Git in Zig called Nit saves AI tokens 71%, speeds up commands, simplifies output, and maintains full functionality.
American forces bombed a school in Iran with human-made targeting errors; AI's role is misdirection from bureaucratic failures.
Anthropic limits Claude usage during peak hours to balance demand and capacity, shifting access and encouraging off-peak activity.
AI bots now surpass human traffic online, growing 8,000% in 2025, transforming internet interactions and replacing human-driven activity.
Anthropic warns users that session limits are faster during peak hours, potentially leading to temporary blocks.
Google's Gemini 3.1 Flash Live enhances audio AI with faster, more natural, multilingual voice interactions and improved complex task handling.
Relay is an open-source, desktop control center for OpenClaw, offering flexible, compliant, and sovereign workflow management for enterprise AI tasks.
Sup AI sets new accuracy on Humanity's Last Exam (52.15%) using ensemble, confidence scoring, extensive search, and advanced context compression.
Paseo enables self-hosted, cross-device, voice-controlled interface for managing Claude, Codex, and OpenCode agents seamlessly.
US midterms spotlight AI's clash with regulation, populist versus institutional views, and growing voter concern over industry influence.
Client-side HNSW vector search in 54KB WebAssembly, enabling sub-millisecond searches without server or API costs.
AI bug reports in Linux shift from junk to genuine, boosting efficiency and review quality amid increased AI adoption in open source.
Layerleak scans Docker Hub images for secrets, analyzes OCI image internals, deduplicates findings, and optionally stores results in Postgres.
On-device AI models like Apple’s Gemini could drive next-generation smartphone upgrades with larger, faster models improving user experience.
OpenAI pauses plans to release an erotic chatbot, focusing instead on core products amid societal concerns and internal strategy shifts.
Context Plugins provide precise API integration directly in IDEs, reducing errors, development time, and costs.
OpenAI shelved plans for an adult chatbot due to concerns over training, legality, investor backlash, and focus on core products.
Wit_coordination minimizes AI merge conflicts by declaring intents, locking symbols, and detecting overlaps before code changes occur.
A custom Lisp-like language improves AI agent safety, efficiency, and debugging by enabling multi-turn, memory-resident, sandboxed code execution on the BEAM.
Rescued my 2013 usability checklist for the AI age; emphasizes localization, transparency, accessibility, performance, privacy, AI transparency, and user well-being.
Breathe-Memory injects associative memory into LLMs, optimizing context retention and reducing token use through structured memory management.
Dependence on AI fuels US growth but risks collapse due to resource shortages, geopolitical conflict, and an looming economic bubble.
MoonshineFlow is a macOS on-device speech recognition app that streams live text into focused apps via the menu bar.
AI whistleblowers reveal safety risks and suppression efforts in major tech firms, inspiring policy changes and public scrutiny.
Apple to open Siri to rival AI assistants in iOS 27, boosting interoperability and user choice.
Open-source proxy enhances AI security by intercepting, evaluating, and logging tool calls with policies, RBAC, and audit trails.
User blocked by network security; file a ticket if the block is mistaken.
Unclear due to access error; unable to determine if agentic AI threatens software engineers.
Cloudflare's Dynamic Workers speed up AI agent code execution by 100x with lightweight, secure isolates, targeting high-volume, short-lived workloads.
Tasker offers rule-based, predictable Android automation; Droidrun uses AI for adaptive, cross-platform workflows and UI control.
Cloneify automates business knowledge queries, workflows, and communication via chat, Slack, and email—saving time and enabling self-sufficiency.
Robotic system recovers and verifies components from retired electronics to reduce downtime and supply constraints.
Visualizes Claude Code agent workflows in real-time, aiding debugging, understanding, and optimizing AI agent orchestration.
AI model identity dominates personality in disagreement patterns; training data influences ratings more than prompts.
OpenAI halts ongoing plans for an erotic chatbot due to misuse concerns, delaying its launch indefinitely.
Easily import your AI chat history and preferences from other apps into Gemini for a seamless, personalized switch.
AI pipeline converts books into graphic novels with AI-generated scripts, art styles, characters, and layouts, producing PDF/EPUB outputs.
Without tolls or charges, autonomous cars risk causing gridlock akin to the Ogallala Aquifer depletion; early regulation is crucial.
Real-time reinforcement learning improves Composer by using live user data for rapid updates, minimizing train-test mismatch, but introduces reward hacking risks.
A sophisticated LiteLLM supply chain attack highlights the need for defense-in-depth AI security layers, including egress control and behavioral monitoring.
AI led police to wrongly arrest a grandmother, highlighting risks of AI misuse in law enforcement.