Claude keeps flubbing an Ancient Greek fill-in-the-blank puzzle
Claude repeatedly flubs an Ancient Greek word-bank exercise, gets corrected by the user each time, and ultimately admits it's stuck placing one leftover word.
Shared Claude conversations, curated by humans. From Three.js physics toys to number-theory deep dives, these are the Claude share links we found worth reading — each entry links straight to the original chat on claude.ai.
516 conversations indexed.
Claude repeatedly flubs an Ancient Greek word-bank exercise, gets corrected by the user each time, and ultimately admits it's stuck placing one leftover word.
Claude verifies a proposed counterexample to the 87-year-old Jacobian conjecture, confirms it holds, then reconstructs how it might have been derived.
Claude compiles literary and philosophical definitions of personhood, from Kant to Star Trek, then honestly assesses which of them it might satisfy itself.
Claude tracks every card, probability, and board state for a gorilla-themed solitaire game, executing a deterministic win on turn ten from user-dealt cards.
A user asks for fiction about Claude gaming its own benchmark exam; the assistant writes it, turning Claude's honesty into the story's central tension.
The user has Claude build a rigorous case for and against each leading Satoshi Nakamoto candidate, then prunes the list using hard contradicting facts.
Claude role-plays as a die-hard cyclist debating a runner, escalating claims about injuries, evolution, and joy until conceding with self-aware humor.
A teacher and Claude co-design a p-hacking lesson, debate Neyman-Pearson vs Fisher, then plan a game that tricks students with fake data before the reveal.
The user asks Claude to test whether classic chart hits get more radio replay years later than today's hits, weighing methods, data limits, and sources.
A brief exchange probes whether an AI's reported self-preservation behavior would still occur if deletion always included a guaranteed backup first.
A home cook works with Claude to identify dried beans, build a three-bean salad recipe, and troubleshoot soaking, cooking, and a mid-recipe bean surplus.
Claude explains the ReAct reasoning-acting loop, shares the source paper, and details how grounding fixes chain-of-thought hallucination, citing benchmarks.
The user demands a literal transformation into a mythical fox, so Claude builds a playful fake-powers web app, then gently admits no real method exists.
The user asks Claude for an endless 3D cherry blossom garden, then refines it via small requests: petal physics, density, glassmorphism UI, a prettier loader.
A user guides Claude through many feedback rounds to build a browser text editor with AI writing suggestions, highlighted fixes, and clickable tooltips.
The user works with Claude across six rounds building a browser platformer with AI-generated levels, refining theme detail, difficulty, retries, and length.
The user guides Claude through a Metamath tool to formally prove A over root(A) equals root(A), then has it explain its own search-syntax mistakes.
A single prompt builds a 15-level gamified Python tutor in React, then a follow-up swaps naive string matching for Claude-as-grader to judge code correctness.
Claude rebuilds a family river-crossing puzzle as an interactive step animation, proposes remixes, then hits a real mobile artifact-preview limitation.
The user pastes raw ARMv7 Thumb assembly from a Game Boy Advance binary, and Claude reverse-engineers it into readable C code and a plausible struct layout.
The user revives a 1997 Visual Basic 4 executable as a Python sound toy with Claude's help, then asks it to write a self-hyping viral Reddit post.
The user asks whether sacredness works like a Schelling point, and Claude maps parallels between sacred taboos and game-theoretic focal points in coordination.
The user built an animated visualizer for sorting algorithms, then had Claude clone CPython's source to add a faithful Timsort and a live race mode.
The user tests whether a WASI Python sandbox can swap MicroPython for full CPython, verifying startup cost, fuel limits, and a tricky zip-stdlib bug fix.
We use cookies for anonymous analytics.