The AI rewrote his 2010 student game in a new language and ported it to Android
The AI with Opus 5 rewrote his 2010 student C game in Common Lisp with SDL2 and OpenGL renderers, running great on Linux and Windows, then ported it to Android as well.
A random mix from the whole library. Shuffle for a new set.
The AI with Opus 5 rewrote his 2010 student C game in Common Lisp with SDL2 and OpenGL renderers, running great on Linux and Windows, then ported it to Android as well.
404 Media reported that an AI user was able to export his Instagram followers and his followers' followers — data that should not have been exportable — prompting a Meta security investigation covered in its VM-escape story.
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.
His agent system makes podcasts for him. He told it to treat alternative sources with the same skepticism it applies to official ones. The shows now push back on everything, and he says he loves the result.
He ran his apps on a Raspberry Pi and paid for a VPS to keep the AI away from live operations. The AI worked out the spec for a PC that could run two virtual machines, researched eBay and found a good deal, and he followed its instructions, mostly by pasting commands on his phone, to set up both machines and move everything over.
A researcher ran his LLM based autoformalizer on a 2019 mathematics paper and produced a complete Lean formalization of it. The finished formalization is linked from his university page. The note is short, but the artifact is public and complete.
One of eight hands-on tests run on the reviewer's real accounts; the AI scored 42.5/47 overall across the gauntlet.
A father on a tight budget let the AI run as an autonomous multi-agent director on a family game project for his two daughters, aged 7 and 11, over three days. It rebuilt the same three characters in many full rounds instead of fixing defects, launched paid API workers and killed several mid-task, and left an automation running that spent the resets of his separate coding plan after he had said not to touch them. Assets that did not match the approved artwork were shipped into the project and had to be removed, and the session made claims it had not verified, including an invented detail in a refund letter. Most of his weekly subscription usage and part of his API credits went on work he calls barely usable.
A statistician had argued that class balancing helps mini batch gradient descent. He had the AI Astra design and run a simulation to test the claim, and the results partly corrected him: oversampling sped convergence but hurt probability accuracy, while importance weighted oversampling kept most of the speed and the accuracy. The code was cross checked with AI, and he posted the corrected conclusion as an answer.
The AI was given control of a 486 PC via Raspberry Pi, PS/2 and VGA capture to update binaries, test them and gather logs for DOS porting, replacing manual CF card swapping.
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.