The Ideabro Brief #5: the murky middle
Models can now read a novel in one go. That's a genuinely impressive feat of engineering, and it's led a lot of us, me included, to assume they remember it the way we would.
They don't. Lena's lead story explains what a context window really is, why material in the middle of a long document tends to get the least attention, and the simple habits that work around it. If you've ever had an assistant confidently describe a table that doesn't exist, this one is for you.
It pairs with a short, sharp lesson in compounding error: why a step that works 95% of the time becomes a coin toss when you chain twenty of them together.
Also this week: an Idea Autopsy of an AI tutor with 180,000 sign-ups and almost no students, and a practical etiquette for saying "I used AI".
— Maya
Models can now read a novel in one go. That doesn't mean they remember it the way you do. Here's what a context window really is, and how to work around its blind spots.
by Lena Hartmann
A 95%-reliable step sounds great until you chain twenty of them together. A quick lesson in compounding error, and how builders fight back.
by Lena Hartmann
An AI homework tutor went viral in September and quiet by March. What killed it wasn't the model. It was the calendar, the parents and a missing definition of 'learning'.
by Theo Brooks
Wedding toasts, condolence notes, product specs: we have powerful new writing tools and almost no shared manners for them. A working guide to when, and how, to disclose.
by Maya Okafor
Uses versus models, rules that expire, sandboxes that aren't loopholes: Theo Brooks talks with a nonpartisan researcher who tries to get both sides of the AI debate in one room.
by Theo Brooks
For long documents: make the model extract before it analyses. It's the single best fix for the murky middle.
From section {{section}} of the attached document ONLY, quote every
sentence that relates to {{topic}}, with page numbers.
Do not summarise or analyse yet. If there are no relevant sentences, say so.
In our policy interview, Ilse Marchetti-Owusu said the most dangerous AI in your town is probably "the boring one making decisions about people that nobody has checked". It's the sentence our team has quoted most this month.
Every weekday
The five AI stories that matter, explained in five minutes.