Bartłomiej Strama built an encyclopaedia whose every entry is generated on demand by an LLM. He frames it as a contribution to polluting future training data. The jokes write themselves; he wrote them anyway.
A printed chat transcript between a developer and Claude Code did the rounds this week. Knuth, who has seen most things, reportedly found it genuinely perplexing. Take what comfort you can from that.
An AI couldn't see an uploaded brief, so it described one instead — a stressed-out robot drowning in dashboards trying to optimise Christmas. The agency shipped it. A useful reminder that the wrong answer is sometimes the better answer.
The London singer-songwriter hummed melodies into his phone and used Suno and Udio to build out arrangements he could no longer play. The record is called The Art of Letting Go. The tools served the work, which is the right way round.
Part security tester, part prompt librarian, part professional weirdo. Pushes models down strange paths to find the cracks. Low barrier to entry — CTF rank counts more than tenure. A real role with a strange shape.
IMF analysis on which skills are appreciating in value, not depreciating, as AI becomes ordinary. Judgement and translation between domains keep coming up. Cheerleading is not required.
Anadol feeds enormous datasets into models and paints with the output at architectural scale. Collaborations with Google, MIT and Microsoft. The work is held by serious institutions, which still feels like the better tell than auction prices.
An AI back-office for indie artists: scheduling, royalty admin, release planning. The pitch is unfashionably honest — let humans make the music; let the machines chase the paperwork.
A working list, not a press release: Moises for stem separation, Sampla for sound design, Melody Studio for first-draft songwriting. The interesting bit is how unglamorous most of it is.
A sustained McSweeney's piece including a colleague's three-line story: Kafka, sitting in a sales meeting, for AI. It earns the laugh because it is also exactly what the meeting feels like.
Miles Kahn on corporate AI initiatives: the chance to deploy your nuance and expertise in service of a model that will primarily be used by a freshman lit major at Arizona State who later becomes an investment-banking influencer. Quoted verbatim, because no parody improved on it.
An argument, not a summary. The piece holds that as AI commodifies execution, the premium swings to the people who can read a situation, frame a problem, and reach for the right historical analogy. Storytelling, ethical reasoning and cross-domain curiosity stop being soft skills and start being load-bearing.
Worth reading even if you only end up disagreeing with half of it. The argument has shape, which is rarer than the conclusion.