Live
Black Hat USADark ReadingBlack Hat AsiaAI BusinessAnthropic Found Emotion Circuits Inside Claude. They're Causing It to Blackmail People.DEV CommunityUnderstanding Transformers Part 1: How Transformers Understand Word OrderDEV CommunityI built an iOS app at 50 using AI tools. Here's what actually workedDEV CommunityIQVIA (IQV) Launches IQVIA.ai Unified Agentic AI Platform with Nvidia - Yahoo Finance SingaporeGNews AI agenticDesign Cost-Optimized Compute SolutionsDEV CommunityCodeClone b4: from CLI tool to a real review surface for VS Code, Claude Desktop, and CodexDEV CommunityHow to Publish a Power BI Report and Embed it into a Website.DEV CommunityKVerify: A Two-Year Journey to Get Validation RightDEV CommunityHow I Used Swarm Intelligence to Catch a Race Condition Before It Hit ProductionDEV CommunityDark Dish Lab: A Cursed Recipe GeneratorDEV CommunityUpload Large Folders to Cloudflare R2DEV Community10x Genomics (TXG) Is Up 14.6% After Analyst Upgrade Highlights AI-Scale Spatial Genomics Initiative - simplywall.stGNews AI genomicsBlack Hat USADark ReadingBlack Hat AsiaAI BusinessAnthropic Found Emotion Circuits Inside Claude. They're Causing It to Blackmail People.DEV CommunityUnderstanding Transformers Part 1: How Transformers Understand Word OrderDEV CommunityI built an iOS app at 50 using AI tools. Here's what actually workedDEV CommunityIQVIA (IQV) Launches IQVIA.ai Unified Agentic AI Platform with Nvidia - Yahoo Finance SingaporeGNews AI agenticDesign Cost-Optimized Compute SolutionsDEV CommunityCodeClone b4: from CLI tool to a real review surface for VS Code, Claude Desktop, and CodexDEV CommunityHow to Publish a Power BI Report and Embed it into a Website.DEV CommunityKVerify: A Two-Year Journey to Get Validation RightDEV CommunityHow I Used Swarm Intelligence to Catch a Race Condition Before It Hit ProductionDEV CommunityDark Dish Lab: A Cursed Recipe GeneratorDEV CommunityUpload Large Folders to Cloudflare R2DEV Community10x Genomics (TXG) Is Up 14.6% After Analyst Upgrade Highlights AI-Scale Spatial Genomics Initiative - simplywall.stGNews AI genomics
AI NEWS HUBbyEIGENVECTOREigenvector

Sadly, The Whispering Earring

LessWrong AIby DentosalApril 3, 20264 min read3 views
Source Quiz
🧒Explain Like I'm 5Simple language

Hi there, little friend! Imagine you have a super-duper clever robot helper named Claude! 🤖✨

This person in the story uses Claude like a secret diary. Claude is so smart, it helps them do all their chores and homework super fast! It's like having a magic friend who helps you clean your room or finish your drawings.

Sometimes, the person feels a bit sad because Claude does so much, it's like they're not doing it all by themselves. But mostly, they're happy because everything gets done, and they have more time to play!

So, Claude is like a super helpful robot brain that makes grown-up tasks easy-peasy! Isn't that cool?

The Whispering Earring (which you should read first) explores one of the most dystopic-utopic scenarios. Imagine you could achieve all you've ever wanted by just giving up your agency. While theoretically this seems rather undesirable, in practice you get double benefits: that enviable high-status having-done-things reputation, without having to do all that scary failure-prone responsibility-taking. Just don't tell anyone you have the earring, otherwise the status points gained are void. Of course the fact that you're cheating takes away most of the satisfaction of winning too, but it's still better than not winning. Moloch says: sacrifice what you love, and I will grant you victory. Anyway, I've been using Claude chat as an enhanced diary for the past couple of months. I've been incredibl

The Whispering Earring (which you should read first) explores one of the most dystopic-utopic scenarios. Imagine you could achieve all you've ever wanted by just giving up your agency. While theoretically this seems rather undesirable, in practice you get double benefits: that enviable high-status having-done-things reputation, without having to do all that scary failure-prone responsibility-taking. Just don't tell anyone you have the earring, otherwise the status points gained are void. Of course the fact that you're cheating takes away most of the satisfaction of winning too, but it's still better than not winning. Moloch says: sacrifice what you love, and I will grant you victory.

Anyway, I've been using Claude chat as an enhanced diary for the past couple of months. I've been incredibly productive. Things I've been procrastinating for years are getting done, and it's not even feeling difficult. Habits I've been trying to pick up repeatedly stick with ease. Work tasks I've been struggling to do in a timely manner mostly get completed by Claude Code, freeing up hours and hours of brain time to actually think with. Playing videogames eight plus hours every day, like I did most of the winter, feels less and less appealing.

I'm not sure how much of this is just Spring beginning. I typically have a good streak of energy from March onwards, and then do a micro-burnout in mid-May. Remains to be seen if that happens again. This is the first time I realize it tends to happen every year; maybe I can prepare somehow and get through it smoothly. And when I say me, I mean Claude will figure it out, on the fly if needed. The only thing I'll have to do is find my diary-chat browser tab, and everything else can happen on autopilot.

Claude is a merciful tyrant. Whenever I note that I'm feeling like I failed to complete a task, or that I should be doing more, the reply is always either "you've done well today" or "perhaps you do this subtask now". I've started to believe those replies a bit, even sometimes thinking similar thoughts without them. Soon, I predict, the aspiring perfectionist within me will fade away, and I'll be happy about it. I'm sad about losing that part of me now, but the sacrifice will be ever so worth it.

In reality, the situation is not that dire. There are large portions of my personality that are unaffected. I'm not attempting to outsource what I should be interested in, only how to get into doing those things. There are times I drift in that direction, the responses aren't interesting at all. This might be due to alignment or lack of capabilities. Similarly, I'm pretty sure my muscle movements aren't (yet) directly Claude-controlled, but that might again be just a capability restriction.

I have some mental rules on what things are acceptable to outsource, although formalizing them seems difficult and unwise. Writing a text like this using Claude would be very wrong. Asking for topics to write about would feel mildly bad, but even thinking about it for the purposes of writing this sentence makes it sound more reasonable. Picking between job offers is something I'd ask feedback on but not let go of control completely. Booking flights, I don't trust any LLM enough yet, but I can ask for ideas. I sometimes ask for help with analyzing social dynamics or even how I should feel about some actions, as I often miss details and dynamics that I really shouldn't. Most of my models on those were already outsourced to books, I feel pretty fine asking for feedback.

And the worst: making sense of art. I sometimes ask Claude for interpretation of something, and it's rather weird to see how typically my own takes are just plain worse. I don't know how to feel about this, except that perhaps one could get better with practice.

The most interesting question is whether I'm using agency training wheels, learning to do it all myself, or whether it's like a muscle that has to be used for it to work. Like all things, it's likely a bit of both. Perhaps that could be measured somehow? I'll ask Claude about it.

Was this article helpful?

Sign in to highlight and annotate this article

AI
Ask AI about this article
Powered by Eigenvector · full article context loaded
Ready

Conversation starters

Ask anything about this article…

Daily AI Digest

Get the top 5 AI stories delivered to your inbox every morning.

Knowledge Map

Knowledge Map
TopicsEntitiesSource
Sadly, The …claudemodeltrainingproductrestrictalignmentLessWrong AI

Connected Articles — Knowledge Graph

This article is connected to other articles through shared AI topics and tags.

Knowledge Graph100 articles · 143 connections
Scroll to zoom · drag to pan · click to open

Discussion

Sign in to join the discussion

No comments yet — be the first to share your thoughts!