惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

V
Vulnerabilities – Threatpost
T
The Blog of Author Tim Ferriss
S
SegmentFault 最新的问题
D
DataBreaches.Net
博客园_首页
罗磊的独立博客
B
Blog
T
Threat Research - Cisco Blogs
C
Cisco Blogs
GbyAI
GbyAI
Engineering at Meta
Engineering at Meta
WordPress大学
WordPress大学
G
GRAHAM CLULEY
H
Help Net Security
酷 壳 – CoolShell
酷 壳 – CoolShell
Cyber Security Advisories - MS-ISAC
Cyber Security Advisories - MS-ISAC
爱范儿
爱范儿
SecWiki News
SecWiki News
T
Threatpost
钛媒体:引领未来商业与生活新知
钛媒体:引领未来商业与生活新知
Schneier on Security
Schneier on Security
T
The Exploit Database - CXSecurity.com
Google Online Security Blog
Google Online Security Blog
T
Tor Project blog
小众软件
小众软件
让小产品的独立变现更简单 - ezindie.com
让小产品的独立变现更简单 - ezindie.com
Y
Y Combinator Blog
H
Hacker News: Front Page
V
V2EX
Security Latest
Security Latest
Cloudbric
Cloudbric
Simon Willison's Weblog
Simon Willison's Weblog
Attack and Defense Labs
Attack and Defense Labs
D
Darknet – Hacking Tools, Hacker News & Cyber Security
P
Proofpoint News Feed
博客园 - 三生石上(FineUI控件)
NISL@THU
NISL@THU
S
Secure Thoughts
Blog — PlanetScale
Blog — PlanetScale
博客园 - 司徒正美
V2EX - 技术
V2EX - 技术
Vercel News
Vercel News
P
Palo Alto Networks Blog
IT之家
IT之家
MyScale Blog
MyScale Blog
有赞技术团队
有赞技术团队
Application and Cybersecurity Blog
Application and Cybersecurity Blog
D
Docker
Google DeepMind News
Google DeepMind News
Webroot Blog
Webroot Blog

Hacker News - Newest: "AI"

AI can't read an investor deck AI as an attorney? Student uses ChatGPT, Gemini to sue UW over alleged racial discrimination Hacking MCP Servers in AI Systems – The Rug Pull: Tool Changes After Approval GitHub - MeepCastana/KubeezCut: Free Web based video editor GitHub - GenAI-Gurus/awesome-eu-ai-act: Curated tools, official sources, OSS, templates, and guides for EU AI Act compliance. Can AI judge journalism? A Thiel-backed startup says yes, even if it risks chilling whistleblowers Coming soon: 10 Things That Matter in AI Right Now DARPA built an AI to fact-check enemy weapons claims What explains heterogeneity in AI adoption? When AI Meets Muscle: Context-Aware Electrical Stimulation Promises a New Way to Guide Human Movements - Department of Computer Science AI Changed How We Build. It Did Not Change What Matters. Linux rules on using AI-generated code - Copilot is OK, but humans must take 'full responsibility for the… Meta spins up AI version of Mark Zuckerberg to engage with employees Code Mode: Let Your AI Write Programs, Not Just Call Tools | TanStack Blog GitHub - Delavalom/graft: Go framework for building AI agents. Type-safe tools, multi-provider (OpenAI, Anthropic, Gemini, Bedrock), zero vendor SDKs. India's TCS tops estimates, says new AI models did not dent services demand Gen Z's fading AI hype Strong feeling: we are in a folded AI reality GitHub - machinarii/total-recall-catalog: A reference catalog of latest knowledge retrieval, memory & RAG systems GitHub - mensfeld/code-on-incus: Give each AI agent its own isolated machine with root, Docker, and systemd. Active defense detects and stops threats automatically.. Quantization, LoRA, and the 8% Problem: Benchmarking Local LLMs for Production AI Iran war: We spoke to the man making Lego-style AI videos that experts say are powerful propaganda Powell, Bessent discussed Anthropic's Mythos AI cyber threat with major U.S. banks GitHub - immartian/bellamem: Persistent belief-graph memory for AI agents. Retrieves decisive context by importance — not recency, not RAG, not /compact. recursive-mode: The Repo-Native Operating System for AI Engineering After the attack on Sam Altman's home, will AI CEO's go on the offensive? The biggest advance in AI since the LLM Opus 4.6 vs GPT 5.4 One Prompt Unity World Generation Test “AI polls” are fake polls Client Challenge Can AI be a 'child of God'? Inside Anthropic's meeting with Christian leaders How to Switch AI Chatbots and Why You Might Want To GitHub - MattMessinger1/agentic_refund_guardrail: Safe refund policy layer for AI agents — Python + TypeScript. Same behavior, shared tests. Adam/papers/emergent_values_whitepaper.md at master · strangeadvancedmarketing/Adam Ask HN: How do you stop playing 20 questions with your AI coding tools How far can automation and AI support psychotherapy? - @theU GitHub - stagas/rtdiff: realtime git diff gui and AI-assisted commits A Mac Studio for Local AI — 6 Months Later A History of the Early Years of AI at the University of Edinburgh Why AI Coding Tools Still Feel Stuck on Localhost MSN AI Datacenters Are Becoming Strategic Targets twitter.com Penn Researchers Use AI to Surface Unreported GLP-1 Side Effects in Reddit Posts Show HN: MoodSense AI (ML and FastAPI and Gradio, Deployed on Hugging Face) Moodsense Ai - a Hugging Face Space by aman179102 AI models are terrible at betting on soccer—especially xAI Grok GitHub - xialeistudio/echoic GitHub - HimashaHerath/github-dev-wrapped: AI-powered weekly GitHub activity reports deployed to GitHub Pages GitHub - alejandrobalderas/claude-code-from-source: Architecture, patterns & internals of Anthropic's AI coding agent — reverse-engineered from source maps AI and Tech brief: Ireland ascendant GitHub - Titovilal/context0: Context0 - Never Surrender Training for a Marathon with an AI Coach: What Worked and What Didn't Cyber Pulse: Agentic Intel - Apps on Google Play I Built an AI PR Reviewer That Catches Bugs by Not Looking for Bugs Gen Z workers are so fearful AI will take their job they’re intentionally sabotaging their company’s AI rollout | Fortune How AI Is Reimagining the Game of Golf–For Both Players and Courses GitHub - nattergabriel/reseed: A CLI tool for managing and distributing agent skills across projects Is SVG the final frontier? My AI workflow evolved from prompts to a near-autonomous workflow MLSharp Help - 3DGS Viewer & Generator I put my cognitive field based AI's runtime on GitHub Is Numble the first AI-proof game? A3: Kubernetes for autonomous AI agent fleets | Emergent Principles Deepali Vyas ("The Elite Recruiter") GitHub - msmarkgu/RelayFreeLLM: A restful API designed to route user prompts to various AI model providers. Unionized ProPublica staff are on strike over AI, layoffs, and wages Unleashing the Advantage of Quantum AI We're heading for an AI-fueled 'dementia crisis,' brain scientist warns The AI-Assisted Breach of Mexico's Government Infrastructure [pdf] GitHub - stef41/lmscan: 🔍 Detect AI-generated text and fingerprint which LLM wrote it. Open-source GPTZero alternative. Zero dependencies, works offline. MSN GitHub - visionscaper/collabmem: Enabling long-term collaboration with Agentic AI - building up episodic and world model memory over time with in-context awareness We gave an AI a 3 year retail lease in SF and asked it to make a profit | Andon Labs AI Code is Hollowing Out Open Source, and Maintainers are Looking the Other Way What leaked "SteamGPT" files could mean for the PC gaming platform's use of AI AI is the boss at this retail store. What could go wrong? GitHub - Wuzu11517/agentic-proxy: Local proxy meant to help reduce With Drones, Geophysics and ArtificiaI Intelligence, Researchers Prepare to Do Battle Against Land Mines A Single Operator, Two AI Platforms, Nine Government Agencies: The Full Technical Report 在 Steam 上购买 FriedrichAI: Offline AI 立省 10% GitHub - inevolin/resume-cli: Hit Claude usage limits? Resume any AI coding session elsewhere. Switch tools at zero friction. GitHub - atripati/ark: AI Runtime Kernel — a context operating system for AI agents. Eliminates tool bloat, loads only what’s needed, and gives LLMs their reasoning space back. How to Build a Secure AI PR Reviewer with Claude, GitHub Actions, and JavaScript This Startup Wants You to Pay Up to Talk With AI Versions of Human Experts Intel Arc Pro B70 Brings 32GB VRAM to Local AI for $949 WordPress 7.0: The Good, the AI, and the Still Missing AI on the couch: Anthropic gives Claude 20 hours of psychiatry IatroBench: Pre-Registered Evidence of Iatrogenic Harm from AI Safety Measures AI Agents Know About Supabase. They Don't Always Use It Right. The history and future of AI at Google, with Sundar Pichai Inside an AI‑enabled device code phishing campaign How Meta Used AI to Map Tribal Knowledge in Large-Scale Data Pipelines AI for Systems: Using LLMs to Optimize Database Query Execution Forecasting the Economic Effects of AI Introducing Tinker: Play with AI, bring your ideas to life AI sheds light on an ancient gaming mystery People really hate AI but not as much as Iran—or Democrats | Fortune What is an AI Product Engineer? Phoebe Gates wants her $185 million AI startup to succeed with 'no ties to my privilege or my last name': 'I have a chip on my shoulder' | Fortune
GitHub - mmarseglia/cognitive-surrender: A short essay on the Wharton finding that users adopt AI answers even when those answers are wrong. Plus, installable Claude prompts that push back. Based on Shaw & Nave (2026) and David McRaney's "You Are Not So Smart".
mmarseglia · 2026-04-24 · via Hacker News - Newest: "AI"

Cognitive surrender: the Wharton paper every AI-coding engineer should read

1. The merge I couldn't re-read

A month ago I merged a TypeScript refactor of my MCP server middleware. Claude wrote it; I clicked through. The code worked. What it didn't do was read cleanly when I came back to extend it — the AI had over-edited a twenty-line function into something I had to re-understand from scratch.

Shaw and Nave call that move cognitive surrender. In their Wharton working paper (https://ssrn.com/abstract=6097646), participants given optional ChatGPT access on a Cognitive Reflection Test adopted whatever the model said — including secretly planted wrong answers. Shaw and Nave found participants' reported confidence rose about 10% regardless of whether the AI-supplied answer was correct; a control group without AI landed at about 50% accuracy. Cognitive offloading is a different shape: running Prettier over a file, or trusting tsc to catch a type error. You keep the reasoning; you delegate the chore. Surrender is what happened in my merge — I did not test the AI's structural judgment against my own, I substituted it for mine. Offloading lightens a load you still understand. Surrender replaces the understanding.

2. What they actually ran

Shaw and Nave used the Cognitive Reflection Test — a short battery of logic problems where the intuitive answer is wrong and a moment of deliberate thought produces the right one. Their design had a control group solve the problems without AI assistance, and a treatment group with optional access to ChatGPT.

The manipulation was the point: Shaw and Nave secretly seeded ChatGPT's responses so that about half the time the AI gave correct answers and half the time it gave wrong ones. Participants didn't know this. The control group, working unaided, landed at roughly 50% accuracy.

Two results define the paper's contribution. First, when the AI was correct, accuracy jumped — participants rode the correct answer to a higher score. Second, and this is the finding that makes the study citable, when the AI was wrong, participants' accuracy didn't merely stay flat or decline modestly. It fell below the no-AI baseline. Unaided performance was 50%; AI-assisted performance with wrong answers was worse than that.

Shaw and Nave also measured confidence. Across both conditions — correct AI answers and wrong ones — participants' reported confidence rose by roughly 10%. The confidence inflation was independent of whether the AI had actually helped. Participants felt more certain whether the AI had guided them right or led them astray.

3. Why willpower is the wrong tool

David McRaney, host of the You Are Not So Smart podcast (YANSS 337, 2026-04-13), reframes the Shaw and Nave result as something older than AI. Confident, fluent, agreeable AI prose does not land on a rational evaluator who weighs it and decides. It lands on evolved machinery that reads signals for agency and trust, and fires a response before deliberation starts. McRaney calls the pattern agentic pareidolia — the perception of a mind behind text that is only producing the shape of one.

The mechanism, from Niko Tinbergen's ethology, is a supernormal releaser: a signal more exaggerated than anything in nature that triggers an evolved response more strongly than the real stimulus. A herring gull chick pecks at its parent's red-spotted bill to beg for food; present the chick with a detached stick painted with an oversized red stripe and it pecks harder at the stick than at a real gull. McRaney's move is to treat LLM output as the stick. Human prose comes with hesitation and mistakes; AI prose is more fluent and more agreeable than any human writer.

Kahneman's dual-process model splits cognition into System 1 (fast intuition) and System 2 (slow deliberation). Shaw and Nave extend the model with a System 3: AI as a layer that does not merely assist the other two but substitutes for them, letting the user skip System 2 entirely. The CRT result is what that substitution looks like under measurement. The releaser fires before deliberation; the adoption happens pre-consciously. "Be more careful" arrives after surrender has landed.

I've noticed the tell: Claude opens with a structural assertion about my code. The LLM doesn't critique, pose clarifying questions, or provide options; it just asserts. That's when the releaser fires, and I stop noticing I'm not the one reasoning. The merge wasn't a hurry problem; I gave up authority and stopped thinking because the output looked good enough. The same shape appears in social media's dark patterns. Infinite-scroll is a variable reward on the dopamine machinery; it's consistent with machinery we don't willfully control. My new habit: no AI-suggested merges without a Five Whys or Socratic walkthrough first.

4. Defensive Prompts

The three sections above made the case. The next part makes it operational. The Defensive Prompts are system prompts engineered against the mechanisms cognitive surrender exploits — confidence inflation, zero friction, ownership confusion, and bypassed deliberation. If you already buy the argument and want something to install or forward, go there directly. If you just finished §3, the prompts are where the reframe becomes a habit.

5. Attribution

A publication about cognitive surrender whose attribution is blurry discredits itself.

  • Shaw and Nave (Wharton). The term cognitive surrender as distinct from cognitive offloading is theirs. So is the CRT experiment: the manipulated ChatGPT answers, the below-baseline accuracy when the AI was wrong, and the ~10% confidence inflation regardless of correctness. The tri-system extension — AI as a System 3 that substitutes for System 2 rather than assists it — is theirs, detailed in the SSRN working paper cited in §1.

  • David McRaney (You Are Not So Smart). Agentic pareidolia is McRaney's coinage. So is applying Tinbergen's supernormal-releaser concept to LLM output — framing AI prose as the exaggerated signal the evolved machinery reads as more trustworthy than any human writer. Episode: YANSS 337 – How to preserve your critical thinking and avoid cognitive surrender when using AI to solve problems, make decisions, and learn new things (2026-04-13).

  • Kahneman. The dual-process model (System 1 / System 2) is Kahneman's. The tri-system extension is not.

  • Author. The §1 engineering image (the MCP middleware merge) is mine, as is the §3 first-person paragraph — the structural-assertion tell, the authority-surrender mechanism, the Five Whys habit. The dark-patterns aside in §3 (the kin-mechanism connection to an interface pattern the reader already recognizes) is mine. I coined "Defensive Prompts" and made the persuasion-layer / reference-layer split. In the §4 four-mechanism list, ownership confusion is Shaw and Nave's term; confidence inflation, zero friction, and bypassed deliberation are my labels for their findings. The general prompt and four specialized prompts are my prior work, republished here. The selection aid on the prompts landing page is my placement choice.