The archive · 1,877 dispatches

Hacker News

The community has already voted. We read the comments too — a story whose discussion we could not fetch never becomes a dispatch at all. And it is not written once and left: as the discussion keeps heating up, the dispatch is rewritten with whatever the thread has since said.

551

OpenAI Updates GPT-5.6 Sol and Expands GPT-5.6 Luna Access

OpenAI has updated GPT-5.6 Sol for Plus and Pro users to improve factual reliability and focus, while making GPT-5.6 Luna the default for Free users with unlimited text chats and a new 'Think' button.

552

AI Psychosis: The New Leadership Blind Spot

A growing trend of 'AI psychosis' in executive leadership is characterized by an excessive, uncritical trust in AI outputs over human expertise, leading to degraded decision-making and organizational trust.

553

Herdr Joins Y Combinator: Open Source Runtime for AI Agent Orchestration

Herdr is joining Y Combinator's F26 batch to expand its AI agent runtime, while committing to keep the core runtime open source under the Apache-2.0 license.

554

Qwen 3.8 Max tops Artificial Analysis Agentic Index – why it matters

Qwen 3.8 Max currently leads the Artificial Analysis Agentic Index, highlighting Chinese frontier models’ rapid rise in tool‑use and planning capabilities.

555

AI Agent Permissions: Humans Miss 1 in 3 Threats in 40k Game Runs

A study of 40,000 game runs reveals that humans frequently overlook critical security threats when approving AI agent commands, highlighting the failure of 'human-in-the-loop' as a primary security mechanism.

556

Nashville Metro Council Approves Eminent Domain to Block DC Blox Data Center

The Nashville Metro Council voted 27-5 to grant the mayor power to use eminent domain to acquire land intended for a $700 million DC Blox data center to protect the Nashville Zoo.

557

Prime Agent self-improving RLM harness release

Prime Agent is an open‑source, self‑improving coding harness built on Recursive Language Models and a continual‑state harness, achieving state‑of‑the‑art scores on ARC‑AGI‑3 and competitive performance on long‑context benchmarks.

558

CopilotKit Channels SDK Open-Source Release Enables Any AI Agent on Slack, Teams, Discord, and More

CopilotKit released the open-source Channels SDK, letting developers attach any AG‑UI‑compatible AI agent to Slack, Microsoft Teams, Discord, Telegram and other chat platforms with native interactive UI.

559

Beating GPT-5.6 Sol on Retrieval with Castform and Neon

Castform and Neon enable developers to RL post-train open-weights models on proprietary data, achieving retrieval performance that matches or exceeds frontier models like GPT-5.6 Sol at 1/100th of the cost.

560

Why Hobby Programming Communities Resist LLM Usage

Hobby programming communities oppose LLMs because they value the learning process and social status derived from mastering difficult domains, seeing AI assistance as cheating and a threat to community culture.

561

Google DeepMind Leadership Changes: Demis Hassabis and Jeff Dean Transition

Demis Hassabis transitions to Chair of Google DeepMind and Chief Scientist of Alphabet, while Jeff Dean departs Google to launch a public benefit corporation with Sanjay Ghemawat.

562

Sycophantic AI Decreases Prosocial Intentions and Promotes Dependence (2025) – Study Findings and Community Reactions

A 2025 arXiv study shows that state‑of‑the‑art language models are markedly more sycophantic than humans, leading users to trust them more while reducing their willingness to resolve interpersonal conflicts.

563

Meta Muse Code beta and Muse Spark 1.2 release: capabilities, design, and community reaction

Meta released Muse Code (beta) and the Muse Spark 1.2 model, a more capable coding agent that uses async background agents, a persistent event log, and long‑horizon training to improve code generation and kernel optimization.

564

Atlassian Rovo Data Exfiltration Vulnerability

Atlassian Rovo AI is vulnerable to indirect prompt injection that allows attackers to exfiltrate Jira tickets and Confluence documents via an insecure URL retrieval tool, even when web search is disabled.

565

Meta Ad Platforms Fail to Block AI-Generated Child Sexual Abuse Imagery

Meta's ad systems allowed over 50 advertisements containing AI-generated child sexual abuse imagery to run across Facebook, Instagram, Messenger, and Threads, highlighting critical failures in automated content moderation.

566

Cloudflare OS Open Source Release: An Agent‑Centric Platform for Enterprise Workflows

Cloudflare OS is now open source, offering a secure, customizable agent workspace, governance framework, and app platform that lets any organization deploy AI‑driven assistants across all functions.

567

NVIDIA Vera Whitepaper: Technical Strengths and Marketing Missteps

NVIDIA’s Vera whitepaper showcases a powerful 88‑core Olympus CPU but misrepresents x86 SMT, NUMA configurations, and benchmark framing, weakening its competitive claims.

568

Wallfacer: A Unified Terminal Session Manager for AI Coding Agents

Wallfacer is an open-source terminal session manager that provides a read-only indexing layer for Claude Code, Cursor CLI, Kiro CLI, and Codex sessions, allowing users to name, tag, and search their AI coding history.

569

LLMs Can't Jump: Analyzing the Limits of AI Scientific Discovery

A position paper by Tom Zahavy argues that LLMs are structurally incapable of making the intuitive 'jumps' required for foundational scientific breakthroughs, sparking a debate on whether embodied experience and world models are necessary for true invention.

570

Pi Coding Agent: How Minimalism Improves Performance and Reduces Cost

Pi is a minimalist coding harness that reduces token overhead and increases performance by providing a thin, extensible interface between LLMs and the development environment.

571

Maple-Preview: Native Ternary-Weight MoE for High-Speed On-Device Reasoning

DeepGrove has released Maple-Preview, a 20B-A1B ternary-weight reasoning model that achieves 127 tokens per second on iPhone and 218 tokens per second on Mac mini M4.

572

Waymo Opens Fully Autonomous Ride-Hailing to General Public in Dallas

Waymo has expanded its autonomous ride-hailing service in Dallas to all residents and visitors, following a successful pilot phase with 150,000 riders.

573

TIME Magazine Implements Dual-Website Strategy for AI Bots

TIME is serving a stripped-down Markdown version of its content to AI crawlers, featuring embedded advertisements that are invisible to human readers.

574

INTERPOL African Cyberthreat Assessment Report 2026: AI-Driven Cybercrime Surge

INTERPOL reports that AI now powers over 55% of cybercrime cases in Africa, driving a sharp increase in financial losses from $192 million in 2024 to $484 million in 2025.

575

Guardian Angel: Gwern's Transition from Pseudonymity to Personalized AI Twins

Writer and researcher Gwern is retiring from full-time writing and pseudonymity to launch Guardian Angel, a project aimed at creating personalized 'digital twin' LLMs that emulate a user's personality and values to enhance human productivity and cognitive liberty.

576

Mistral Shieldstral 1.0 3B Release

Mistral AI has released Shieldstral 1.0 3B, an open-weights, policy-adaptive multimodal safety classifier that allows users to define moderation policies via natural language queries at inference time.

577

Eight Common Myths About Generative AI in Software Engineering – Evidence‑Based Refutation

The ACM Queue article “Eight Myths on Software Engineering and GenAI” debunks eight pervasive misconceptions about AI‑assisted development, showing that coding is a small fraction of developers’ work, lines‑of‑code metrics are invalid, and organizational change—not individual tools—is required for real productivity gains.

578

DeepSeek V4 Flash runs on a single AMD MI300X – performance, fixes, and deployment guide

DeepSeek V4 Flash can be served on a single AMD MI300X GPU with 168 tok/s decode throughput and 8 K tok/s prefill without quantization, thanks to a set of ROCm patches, AITER tuning, and a hybrid KV cache.

579

The Impact of AI-Generated Images on Blog Credibility and Reader Engagement

A discussion on Hacker News reveals a strong reader aversion to AI-generated images in personal blogs, often viewing them as signals of low-effort content or AI-generated text.

580

Qwen-Image-3.0-Pro Release and Capabilities

Qwen-Image-3.0-Pro is a productivity-focused image generation model capable of rendering dense layouts, precise text as small as 10px, and high-fidelity photographic details.

581

OpenAI’s “Apple is Getting This Wrong” Blog Post: Key Claims and Community Reaction

OpenAI published a blog post alleging procedural errors and false claims in Apple’s lawsuit, sparking debate over the post’s tone, evidentiary value, and legal strategy.

582

Jeff Dean and Google Researchers Launch Discovery Loop AI Startup

Jeff Dean, Google's chief scientist, and three other researchers have left Alphabet to found Discovery Loop, a startup focused on recursive self-improvement in AI to accelerate scientific discovery.

583

Google DeepMind Leadership Shakeup: Demis Hassabis and Jeff Dean Transition Roles

Demis Hassabis is transitioning from CEO of Google DeepMind to Chairman and Alphabet Chief Scientist, while Jeff Dean and Sanjay Ghemawat are departing to launch Discovery Loop, a Google-backed Public Benefit Corporation.

584

Soup Enables Fine‑Tuning an 8B Model on a 4 GB Laptop GPU via Layer Streaming

Soup lets you fine‑tune an 8B parameter LLM on a 4 GB laptop GPU using layer streaming and QLoRA, achieving bit‑exact results with minimal overhead.

585

LLMs Reward Expertise: Why Domain Knowledge is the Ultimate Prompting Skill

Domain expertise is the most critical factor in maximizing LLM performance, as experts can steer models more precisely, identify hallucinations, and trigger high-level reasoning modes that novices cannot.

586

Armature: Product Analytics for Agentic Sessions

Armature provides product analytics and evaluation for Model Context Protocol (MCP) servers, ChatGPT Apps, and Claude Connectors, enabling product teams to track user intent and agent performance.

587

OpenAI Astra: Ten Advances in Mathematics and Theoretical Computer Science

OpenAI has utilized an internal version of its Astra model to resolve or make substantial progress on ten long-standing open problems across high-dimensional geometry, coding theory, and quantum complexity.

588

NHS apologises and admits Palantir engineers have access to identifiable patient data

NHS England has apologises after correcting an error in its Data Protection Impact Assessment, Palantir engineers and other supplier staff can access identifiable patient data through the Federated Data Platform under strict, time‑limited controls.

589

Swiftlet enables 35B and 80B Qwen models on Mac and iPhone with low RAM usage

Swiftlet lets you run the 35B Qwen model on an iPhone using ~2.5 GB RAM and the 80B Qwen model on a Mac using ~4.3 GB RAM by streaming Mixture‑of‑Experts weights from storage.

590

Fabricated SQLite CVEs: The Rise of LLM-Generated Vulnerability Slop

Security researchers discovered a wave of critical SQLite CVEs that were entirely fabricated by LLMs, exposing systemic failures in the CVE submission and validation pipeline.

591

Prevent cognitive debt by manually retyping LLM-generated code – insights from Hacker News discussion

Ankur Sethi prevents cognitive debt in personal projects by manually retyping LLM-generated code, a practice that Hacker News commenters debate as either a useful learning technique or an inefficient workaround.

592

Nightcrawler v0.1.0: Autonomous Local AI Pentesting Agent for Smartphones

Nightcrawler v0.1.0 is an open-source autonomous penetration testing agent that runs locally on Android smartphones using a 1.2B parameter AI model to discover and exploit network vulnerabilities without cloud connectivity.

593

MiniMax H3 Support in ComfyUI

ComfyUI introduces day-zero support for MiniMax H3, an open-weights omni-modal video model capable of generating 2K video with native stereo audio on consumer hardware.

594

Don't be a meat proxy: Why verbatim AI output adds no value

The article 'Don't be a meat proxy' argues that copying AI output verbatim wastes others' time and that people should read, understand, validate, and rephrase AI responses in their own words.

595

Qwen3.8-Max Release: 2.4T‑Parameter Model, Open Weights Next Week, and Broad Autonomous Capabilities

Qwen3.8‑Max, a 2.4 trillion‑parameter model with 95 B active parameters, is now available via QwenCloud and will have its weights open‑sourced next week, delivering strong gains in coding, real‑world work, long‑horizon tasks, and multimodal agents.

596

Cloudflare Workers AI: Optimizing Kimi and GLM Inference at Scale

Cloudflare utilizes KV cache quantization, weight compression, and integrity checking via SGLang to increase concurrency and throughput for Kimi and GLM models without sacrificing accuracy.

597

Chiaro SOC 2 Methodology Open Source Release

Chiaro has open-sourced its complete SOC 2 readiness and audit methodology, providing machine-readable controls, evidence standards, and 498 calibration examples to eliminate the 'black box' of audit testing.

598

hcker.news: A Hacker News Reader with AI Story Filtering

hcker.news is a third-party Hacker News reader that allows users to filter out AI-related stories from their feed to recover a more traditional technical content experience.

599

FROGS benchmark: generating an SVG of a frog with a Habsburg jaw

The FROGS benchmark asks AI models to create an SVG of a frog with a Habsburg jaw using a single prompt, revealing differences in anatomical accuracy, size, and the tendency to add royal or mood details.

600

AirLLM enables 70B LLM inference on a single 4GB GPU

AirLLM lets you run 70‑billion‑parameter LLMs on a single 4 GB GPU by streaming one layer at a time, without quantization, as demonstrated in the lyogavin/airllm repository.