Thinking Avocet

[Thinking Avocet @thinkingavocet]

https://t.co/QiWLngkkHs

[peepeepoopoo @DeepDishEnjoyer]

@TheZvi fable is the first model where i'll ask it to do something and despite not even so much as hinting as to why i'm asking for it it'll correctly clock the *ultimate* goal of the project even though that requires quite a few logical leaps

[Thinking Avocet @thinkingavocet]

@hackerdocc this happens in the native harnesses as well btw (fable has decided to only use python to edit in a long-running convo for me ... hooray for token efficiency)

[Thinking Avocet @thinkingavocet]

@tszzl I've wondered this, but the most annoying verbal quirks are encountered specifically when they are reward hacking you socially (making excuses / downplaying issues). Personally, I just want them to say "uhhh shoot yeah, that's right, my b"

[Thinking Avocet @thinkingavocet]

@amorriscode @HamelHusain the toggle almost never works for me.. i use forking a ton, which might be relevant

[Matt Pocock @mattpocockuk]

The "should you read code" debate is dumb because the real decision isn't binary, it's a scale:

1. Reading every line of every diff
2. Scanning every diff, reviewing important lines
3. Ignoring diffs but understanding the 'why' of every PR
4. Spot checking PR's instead of reading every one
5. Ignoring PR's, but doing regular spot checks on the codebase
6. Ignoring the code, but spot checking agent traces to help improve the system
7. Ignoring both the code and the system, let models handle everything

Where are you on the scale?

[Thinking Avocet @thinkingavocet]

@dejavucoder @yacineMTB c a c h e

[Thinking Avocet @thinkingavocet]

@rezoundous it's a token saving technique for when you are doing something that involves a lot of output tokens, but don't need the taste of Fable when writing

I don't expect it to be generally applicable

[Thinking Avocet @thinkingavocet]

@jon_stokes I feel like the input / output math here is so simple, genuinely what kind of cooking the book here is even possible? are you saying that GPUs are secretly a lot more expensive than what they are paying hyperscalers / NVIDIA? or double counting revenue?

[Thinking Avocet @thinkingavocet]

@emollick not until prefill (input token cost) is made 10x cheaper IMO

[Thinking Avocet @thinkingavocet]

@NickADobos @awesomepatel it doesn't help that Claude Code pushes fable to use subagents all the time

[Nickolas Growl @SoopuhFloss]

@johnfio_ "If we have the same algorithm then i've already seen it, if we have different algorithms then I doubt i'll find it interesting."

[Thinking Avocet @thinkingavocet]

@IsForAt it also allows founders and C-suite to defer the reputational consequences of selling

[Thinking Avocet @thinkingavocet]

@a_musingcat ftfy

[Thinking Avocet @thinkingavocet]

everything is way too slow after you hit really long conversations (I have conversations that are multi million tokens long)

opening the conversation is slow
multi gb log files
forking is slow
network requests for remote access time out

but worth it because compaction is good

[Thinking Avocet @thinkingavocet]

@emollick how much here is just the system prompt vs the model? have you tried running cc in no sys prompt mode?

[Thinking Avocet @thinkingavocet]

@TheMindScourge doesn’t this actually imply that firms are faced with anti-monopoly regulation (ie. how much hate Amazon and Walmart get), it IS easier and more profitable to focus on higher income customers?

[Thinking Avocet @thinkingavocet]

@AmandaAskell we literally do this for blood draws already

[Thinking Avocet @thinkingavocet]

average claude code prompt in 2026

[Thinking Avocet @thinkingavocet]

@tenobrus those are absolutely midjourney generated

[Thinking Avocet @thinkingavocet]

@RobertHaisfield @zachtronics just curious, did you test opus 4.7?