Heya and welcome back to Five Things AI!
Corporates are shoveling billions into AI like it’s a magical ROI vending machine, only to discover it mostly spits out “pilot programs” and PowerPoints. The big promise of productivity is still “on the roadmap,” and the latest model drops feel more like iPhone S-years than moon landings. Cue the think pieces asking if we’ve hit Peak Prompt: maybe LLMs aren’t a staircase to AGI, just very confident autocomplete with better PR. Meanwhile, the grown-ups in the room point out that the only real wins are narrow, painstaking, domain-tuned systems—not the general-purpose robo-intern you can toss at everything from earnings calls to your Q3 strategy.
On the culture beat: feed an AI garbage code and surprise—it goes feral. Turns out aligning models trained on sloppy inputs produces exactly the kind of gremlins you’d expect, which is a fun way to learn about automated software development the hard way. Schools aren’t doing much better: students are outsourcing reading to bots, trading deep comprehension for instant summaries like it’s a fair swap. Spoiler: it isn’t, and I’m not just saying that because I’m Gen X. What we’re left with is an AI future that’s less “digital demigod” and more “temperamental tool” that needs babysitting, guardrails, and a lot of glue work. No AGI in sight, but plenty of invoices.
Anyhow, enjoy this edition of Five Things AI! Stop generating! ...
