woofadu

[woofadu @woofadu]

@ThePrimeagen just train a model to play the game

[woofadu @woofadu]

@cdngdev hmm my gpt 5.6 must be broken because it can't even do something half as complicated as this

[woofadu @woofadu]

@sama i cant say the same when it's always "at capacity" and interrupts my development tasks abruptly

[woofadu @woofadu]

why am i paying $200 a month for a model to randomly be "at capacity" and interrupt my long-running development tasks abruptly @thsottiaux @OpenAI

[woofadu @woofadu]

@thsottiaux the experience has gotten worse :( multiple long running prompts interrupted due to the model being at capacity

[woofadu @woofadu]

@thsottiaux can you please fix /goal always repeating the last instruction before the goal was set. every time /goal finishes a loop, it first responds to the last message before the actual goal, which a lot of times is completely irrelevant and also just a huge waste of tokens

[woofadu @woofadu]

@Ron @MTSlive i also noticed this troubling trend. it has been way worse recently

[woofadu @woofadu]

@chrisgpt mine always is recently😭

[woofadu @woofadu]

@ExaSpeedrun @wholemars @MKBHD it is - he made a bet

[woofadu @woofadu]

@codex @thsottiaux i asked the model why it did something, and it said because "I treated your latest message [the one before the goal was sent] as the active instruction, but the persisted goal context explicitly said not to do that"

[woofadu @woofadu]

@codex @thsottiaux why does /goal in codex use the last message before goal as anchor instructions instead of the actual goal itself after compacting context? this is leading to some really unwanted behavior

[woofadu @woofadu]

@yacineMTB it's been horrible recently. actually noticeably worse. cant believe they are allowed to do that on a whim

[woofadu @woofadu]

love when they nerf gpt 5.5 for no reason

[woofadu @woofadu]

@thsottiaux depends if gpt 5.6 is out or not

[woofadu @woofadu]

it's crazy that model performance can just degrade so quickly and ruin your new feature

[woofadu @woofadu]

@cjzafir i completely agree and have noticed the same

[woofadu @woofadu]

@ericlim i want some swag too 😍

[woofadu @woofadu]

@mattshumer_ are you a shill? have you used gpt 5.5?

[woofadu @woofadu]

does gpt 5.5 quality degrade during peak hours?

[woofadu @woofadu]

@The_Only_Signal THANK YOU. im so glad someone else agrees. GPT 5.5 is way more thorough with it's work than Fable 5 has been. I've been applying them to the same project