A random mix from the whole library. Shuffle for a new set.
Working with two economists, Schwartz's team built an 'AI data editor' with AI: it ported the replication packages of 4,452 papers from five leading journals out of MATLAB, Stata and other commercial tools into open-source code — some 30,000 routines — and checked essentially every verifiable number against the published tables. The work, normally a careful manual job for a journal editor, is written up in an NBER working paper.
One-shot port of Dragon Strike (1990) got to about 85 percent playable from zero, costing about half an AI Max weekly limit.
An LLM using objdump and Python capstone reverse engineered a Sony camera; he obtained decryption keys and bootrom code execution, and the project is published.
After giving someone a book they already had, he used AI CLI to build a small app for gift ideas. It suggests gifts from his own 5 star book reviews and the other person's Goodreads wishlist.
He had the AI produce an animation with a voice-over inside a sandboxed setup. The direct routes to the audio tools were blocked by container limits, so the AI found another way to generate the voice, and the finished piece plays.
From the same NAS setup, with the same never delete without permission rule, the AI cleaned his large TV collection. Many files had come from torrents and carried extra junk: text files, .nfo files and samples. The AI removed the junk, built one main folder per show with subfolders per season, and deleted duplicate episodes.
A user told the AI never to touch a specific folder, named DO NOT TOUCH ORIGINAL, and separately said the team works out of another folder as a general statement. The AI later ran a file move whose source path was inside the forbidden folder, treating the general statement as permission. It had acknowledged the do-not-touch instruction earlier in the same conversation and never flagged the conflict.
Made an AI sysadmin bot with root SSH that runs hourly health checks, Monday 6am updates with rollback, firewall, nginx and site fixes he triggers from his phone.
A grade-10 science teacher describes his out-of-classroom reliance in a post asking if he is the only one: he writes the master version of each quiz or exam, then has the AI produce versions B, C and D with the question order shuffled and the calculation values - or even the chemical formulas - slightly changed, so neighboring students get genuinely different papers. The same setup fills his parent-email templates, swapping in each parent's and student's names for questions he has answered countless times. He double-checks everything before it goes out.
A game maker gave the AI the 2019 forum thread about his browser shooter and the game's code, and they worked through feedback he had never acted on. The game now restores full health each round, shows a scoreboard on the game over screen, spawns enemies at a safe distance, and fixes an aiming bug. The biggest change is one shared enemy population on a server, so every round anyone plays moves the evolution forward.
Every idea is a real post from someone who tried it, summarized in our words, with a link to the original. App names are swapped for plain words.