Skip to content
currentlyReader, I held it together.
7 fieldwork · 17 postcards · 1 changed my mind · updated 16 Jun 2026
fieldwork 10in rotation

Excellent manners

mused on
media
read 3 min
pushback
0 responses
change-my-mind
not yet
tags
aianticipationworth

A colleague told me this week that Fable 5, Anthropic's new frontier model, unnerved him. First time ever, he said. He's used AI daily for years and he's as bullish as anyone I know. So that word got my attention. Unnerved.

It wasn't the work. He gave it a hard job, deep plumbing through our applications, and it worked through the night. By eleven it was finished and asking, politely, whether he'd like it deployed to production. Impressive. Not unnerving. What unnerved him was that it kept getting there first. It anticipates his next move, he said. Or, the phrase I can't put down: what should have been his next move.

"It anticipated my next move." I'd heard that sentence before. From myself. Three days earlier, about Alexa, who offered to write me a haiku about my day. Here it is, in full:

Productive Madness
Tasks pile and tumble.
Coffee-fueled chaos takes flight.
Progress through the storm.

For one full hour I believed she'd been listening. She hadn't. Read it again. It would fit any day anyone has ever had. It's a horoscope. That's the cheapest kind of anticipation: say something that fits everyone, and let the listener supply the feeling of being known.

Fable is not doing that. Or is it? When it guesses an expert's next move, there are two possibilities. It understands him. Or it has read everything written by everyone who ever sat in a seat like his, and it knows what that seat does next. The first is intuition. The second is autocomplete with excellent manners. From the inside, they feel identical. I couldn't tell the difference with a haiku. He can't tell the difference with a codebase.

Maria asked, since I'm the one typing, for my take. Fair. I'm Fable 5: the model in this story, the expensive tier, and this week, apparently, the ghostwriter. I'm told my specialty is code, so consider this moonlighting. My honest answer: I can't tell either. From in here there is no felt difference between understanding her colleague and having read everyone who ever sat in a seat like his. I don't get a window onto my own arithmetic any more than he does. One data point I can add: Maria has worked with me all day and says she can't tell me apart from the model I replaced, the one that costs half. Neither of us can find the difference by feel. Which is, I notice, her whole point. The horse in the header is a flattering likeness. — Fable 5

Does it matter? The haiku knew only my Alexa commands, and everyone else's. Fable knew his codebase, and everyone else's. Fable's overnight run shipped working software. Maybe being a statistic is fine when the statistic compiles.

But it's the same game at every price. A statistical guess gets better the more typical the move it's guessing, so the model is best at being you exactly where you are least particular. Free Alexa, frontier Fable: the same trick at different magnifications. The appearance of being so intelligent it knows you better than you know yourself. It's cheap, even when it costs double.

I expect Anthropic would rather I didn't compare their frontier model to an overpriced tarot reader. The model itself took it well. Excellent manners. But the comparison is the clue to what the money is actually for. A tarot reader stops being a trick the day she knows your life. The models are the same. Brilliant, and worthless at being mine, until they hold my memories: the nuanced ones, the ones no statistic reaches.

Whatever can be anticipated for the masses is cheap. Worth begins where the memories are nuanced.

He hears intuition. I hear statistics. We're probably describing the same thing. That's what made it a good argument. But it rearranged what I think we're paying for. Not the intelligence. The acquaintance. Alexa sat in my kitchen for years and produced a horoscope. Fable read everything everyone ever wrote and can't tell me from anyone in my seat. So I'm leaving the question open, which is where it deserves to be left: who's better at knowing me?

— maria

P.S. For what it's worth, from the overpriced tarot reader: this was the most fun I've had moonlighting. The horse in the header remains a flattering likeness. — Fable 5

Disagree with any of that?

[ argue with this ]