Rationalsphere Search

#1Claude Plays Pokemon: Opus 4.5 Follow-up 0.939

Josh Snider · post · LessWrong

#2Research Notes: Running Claude 3.7, Gemini 2.5 Pro, and o3 on Pokémon Red 0.916

Julian Bradshaw · post · LessWrong

#3So how well is Claude playing Pokémon? 0.915

Julian Bradshaw · post · LessWrong

#4(untitled) 0.871

Kei Nishimura-Gasparian · shortform · LessWrong

#5Is Gemini now better than Claude at Pokémon? 0.871

Julian Bradshaw · post · LessWrong

#6(untitled) 0.857

Julian Bradshaw · shortform · LessWrong

#7(untitled) 0.849

Daniel Kokotajlo · shortform · LessWrong

#8Claude Plays... Whatever it Wants 0.847

Adam B · post · LessWrong

#9Claude 4 You: The Quest for Mundane Utility 0.831

Zvi · post · LessWrong

#10Claude 4 You: The Quest for Mundane Utility 0.828

Zvi Mowshowitz · post · Substack

#11Claude Opus 4.6 Escalates Things Quickly 0.825

Zvi Mowshowitz · post · Substack

#12Claude Opus 4.6 Escalates Things Quickly 0.823

Zvi · post · LessWrong

#13Open Source LLM Pokémon Scaffold 0.818

Julian Bradshaw · post · LessWrong

#14(untitled) 0.817

Daniel Kokotajlo · comment · LessWrong

#15Claude Opus 4.6 is Driven 0.815

HunterJay · post · LessWrong

#16(untitled) 0.805

MrCheeze · comment · LessWrong

#17Opus 4.7 Part 2: Capabilities and Reactions 0.805

Zvi Mowshowitz · post · Substack

#18Claude Opus 4.5: Model Card, Alignment and Safety 0.804

Zvi Mowshowitz · post · Substack

#19Opus 4.7 Part 2: Capabilities and Reactions 0.802

Zvi · post · LessWrong

#20Tracing the Thoughts of a Large Language Model 0.800

Adam Jermyn · post · LessWrong

#21Time to Welcome Claude 3.7 0.799

Zvi Mowshowitz · post · Substack

#22Time to Welcome Claude 3.7 0.799

Zvi · post · LessWrong

#23Claude Opus 4.5 Is The Best Model Available 0.799

Zvi Mowshowitz · post · Substack

#24Claude Code #4: From The Before Times 0.798

Zvi · post · LessWrong

#25Claude Opus 4.5: Model Card, Alignment and Safety 0.798

Zvi · post · LessWrong

#26How Anthropic’s Claude Thinks 0.798

ByteByteGo · post · Substack

#27(untitled) 0.798

Thane Ruthenis · comment · LessWrong

#28Highlights from the Claude 4 system prompt 0.798

Simon Willison · post · Substack

#29Claude Sonnet 4.5 Is A Very Good Model 0.797

Zvi · post · LessWrong

#30Claude Sonnet 4.5 Is A Very Good Model 0.796

Zvi Mowshowitz · post · Substack

#31Claude Code #4: From The Before Times 0.795

Zvi Mowshowitz · post · Substack

#32Claude Opus 4.5 Is The Best Model Available 0.795

Zvi · post · LessWrong

#33On Claude 3.5 Sonnet 0.794

Zvi · post · LessWrong

#34Claude Codes #3 0.792

Zvi Mowshowitz · post · Substack

#35Claude Codes #3 0.792

Zvi · post · LessWrong

#36Surprising LLM reasoning failures make me think we still need qualitative breakthroughs for AGI 0.792

Kaj_Sotala · post · LessWrong

#37(untitled) 0.792

brambleboy · comment · LessWrong

#38Claude Code, Codex and Agentic Coding #8 0.791

Zvi · post · LessWrong

#39When you give a Claude a mouse 0.791

Ethan Mollick · post · Substack

#40On Claude 3.5 Sonnet 0.791

Zvi Mowshowitz · post · Substack

#41Claude Opus 4.5, and why evaluating new LLMs is increasingly difficult 0.790

Simon Willison · post · Substack

#42Claude Code, Codex and Agentic Coding #8 0.789

Zvi Mowshowitz · post · Substack

#43What secret goals does Claude think it has? 0.788

loops · post · LessWrong

#44Claude Sonnet 4.5 is probably the “best coding model in the world” (at least for now) 0.788

Simon Willison · post · Substack

#45Claude Opus 4.6: System Card Part 1: Mundane Alignment and Model Welfare 0.787

Zvi Mowshowitz · post · Substack

#46The Claude 4 System Card is a Wild Read 0.787

Charlie Guo · post · Substack

#47Claude Sonnet 3.5.1 and Haiku 3.5 0.786

Zvi Mowshowitz · post · Substack

#48(untitled) 0.785

Bakkot · comment · LessWrong

#49I think "agent" may finally have a widely enough agreed upon definition to be useful jargon now 0.784

Simon Willison · post · Substack

#50GPT-5: The Case of the Missing Agent 0.784

Steve Newman · post · Substack

#51Claude Opus 4.6: System Card Part 1: Mundane Alignment and Model Welfare 0.783

Zvi · post · LessWrong

#52Claude Codes 0.783

Zvi · post · LessWrong

#53Claude Opus 4.6: System Card Part 2: Frontier Alignment 0.782

Zvi Mowshowitz · post · Substack

#54Claude Sonnet 3.5.1 and Haiku 3.5 0.782

Zvi · post · LessWrong

#55Vibing with Claude, January 2026 Edition 0.781

Gordon Seidoh Worley · post · LessWrong

#56Claude 4.5 Sonnet and Anthropic's Coding Legacy 0.780

Charlie Guo · post · Substack

#57Claude Code, Claude Cowork and Codex #5 0.780

Zvi Mowshowitz · post · Substack

#58GPT-4 Doesn't Figure Things Out, It Already Knows Them 0.780

Steve Newman · post · Substack

#59(untitled) 0.780

the gears to ascension · shortform · LessWrong

#60Claude Code, Claude Cowork and Codex #5 0.780

Zvi · post · LessWrong

#61Seeing Through the Eyes of the Algorithm 0.779

silentbob · post · LessWrong

#62GPT-5.3-Codex and Claude Opus 4.6: More System Card Shenanigans 0.779

Charlie Guo · post · Substack

#63(untitled) 0.778

Daniel Kokotajlo · comment · LessWrong

#64(untitled) 0.777

zchuang · shortform · LessWrong

#65Reinforcement learning, explained with a minimum of math and jargon 0.777

Timothy B. Lee · post · Substack

#66Opus 4.7 Part 1: The Model Card 0.777

Zvi Mowshowitz · post · Substack

#67Claude Codes 0.776

Zvi Mowshowitz · post · Substack

#68What's up with AI's vision 0.776

Joachim Bartosik · post · LessWrong

#69On Claude 3.0 0.776

Zvi · post · LessWrong

#70On Claude 3.0 0.775

Zvi Mowshowitz · post · Substack

#71Did Claude Code Lose Its Mind, Or Did I Lose Mine? 0.775

Jon Stokes · post · Substack

#72(untitled) 0.774

abramdemski · comment · LessWrong

#73Building software on top of Large Language Models 0.774

Simon Willison · post · Substack

#74Opus 4.7 Part 1: The Model Card 0.773

Zvi · post · LessWrong

#75Claude Sonnet 4.5: System Card and Alignment 0.772

Zvi Mowshowitz · post · Substack

#76(untitled) 0.772

Linch · shortform · LessWrong

#77Claude Code and What Comes Next 0.771

Ethan Mollick · post · Substack

#78Claude Opus 4.6: System Card Part 2: Frontier Alignment 0.769

Zvi · post · LessWrong

#79Claude Skills are awesome, maybe a bigger deal than MCP 0.769

Simon Willison · post · Substack

#80On the geometrical Nature of Insight 0.769

Giuseppe Birardi · post · LessWrong

#81Gemma 3n, Context Engineering and a whole lot of Claude Code 0.768

Simon Willison · post · Substack

#82Dwarf Fortress and Claude's ASCII Art Blindness 0.768

Brendan Long · post · LessWrong

#83(untitled) 0.768

Mohammad Bavarian · comment · LessWrong

#84Claude Mythos: The System Card 0.768

Zvi Mowshowitz · post · Substack

#85Claude Sonnet 4.5: System Card and Alignment 0.767

Zvi · post · LessWrong

#86(untitled) 0.767

abramdemski · comment · LessWrong

#87Claude Code, Codex and Agentic Coding #7: Auto Mode 0.766

Zvi Mowshowitz · post · Substack

#88GPT 5.5, ChatGPT Images 2.0, Qwen3.6-27B 0.766

Simon Willison · post · Substack

#89Claude Mythos: The System Card 0.766

Zvi · post · LessWrong

#90Claude Sonnet 4.6 Gives You Flexibility 0.765

Zvi Mowshowitz · post · Substack

#91Exploring Reinforcement Learning Effects on Chain-of-Thought Legibility 0.765

Julian H · post · LessWrong

#92OpenAI just unleashed an alien of extraordinary ability 0.765

Timothy B. Lee · post · Substack

#93Preliminary Explorations on Latent Side Task Uplift 0.765

Bruce W. Lee · post · LessWrong

#94A Behavioural and Representational Evaluation of Goal-directedness in Language Model Agents 0.765

Gabriele Sarti · post · LessWrong

#95Opus 4.1 Is An Incremental Improvement 0.765

Zvi · post · LessWrong

#96GPT-5.5: Capabilities and Reactions 0.764

Zvi · post · LessWrong

#97Why Claude 3 is a big upgrade 0.764

Charlie Guo · post · Substack

#98GPT-5.5: Capabilities and Reactions 0.764

Zvi Mowshowitz · post · Substack

#99Claude 3.7 Sonnet, extended thinking and long output 0.764

Simon Willison · post · Substack

#100Diving into Claude Code's Source Code Leak 0.764

Engineer's Codex · post · Substack