Verification overhead increases token consumption and offsets efficiency gains.
spent the day with fable on a bunch of random stuff and it’s very spiky, imo.
it is brilliant in the same message where it makes 1-3 severe mistakes, which means you have to check even more stuff more in-depth, wasting time and tokens
Positive users highlight helpful exchanges in Fable AI discussions while negative users criticize its spiky outputs, catastrophic errors, dense jargon, and token-wasting design.
@dejavucoder well there’s no pro in codex, but pro is an insanely good model that barely makes mistakes
Atleast tarvangian gets more compassionate when he gets dumb.
so dario seeked out the nightwatcher spren's curse and boon onto mythos/fable, we have tarvangian from first principles
with codex, i feel like the message is either correct or so bad that it’s very obvious if you know what you are doing.
this could be me being more used to gpt, I haven’t used Claude in ~7-8 months as my main model now
@xeophon Empathy is an extremely underrated skill in the agentic era for this reason. I have a very good sense of when Claude will or has gotten something wrong. I code entirely in CC with no editor/IDE and I can just feel when it's done smth stupid
@xeophon agree
@xeophon how do you think it compares to gpt 5.5 pro
@xeophon i keep saying this
@xeophon @dejavucoder 5.5-Pro is lowkey Mythos-class and easily beats Opus, but it rarely gets evaled or really any usage at all because its so expensive, which is a problem they gotta fix.
@xeophon I enjoyed how Fable talked crap about my codebase as if its sibling Opus hasn’t written most of it.
@xeophon yeah I noticed this too
@xeophon i think this is context distillation as opposed to steering vectors. it tries to maintain coherence but is forced on a different trajectory and they clash so you get this blend of solid thinking with assistant mush instead of a/b
@xeophon Have you found that its text outputs are dense and full of nonstandard jargon? that's been my biggest gripe with it so far
@xeophon They want you to burn as many tokens as possible doing repetitive things
That’s their business model
@code_star TVKE I was expecting someone to take the hint on what I meant as well god you're the goat 🐐
@aniketthh @xeophon wouldnt this just be its more jagged?
@xeophon yeah the spikiness is the actual tax. a model that's reliably mediocre you can plan around. one that's brilliant then catastrophic in the same reply means you never get to lower the review effort, no matter how high the ceiling goes
@code_star tarvangian getting softer the dumber he gets is actually kinda sweet in a tragic way
the boon really said compassion is the price of ignorance