Monkey pong clip used to argue old rules should be dropped for GPT-6 Astra
The author uses 2021 footage of a brain-computer-interface monkey still gripping an unplugged joystick to liken developers' redundant rule files for Astra to compensation that has become a defect.
A MONKEY IS PLAYING PONG WITH HIS MIND AND THE JOYSTICK IN HIS HANDS IS NOT PLUGGED INTO ANYTHING his name is Pager. nine year old macaque. two chips, one in each side of his brain, six weeks old at the time of filming. the fur hasn't fully grown back yet he learned the game for a banana smoothie through a straw how it actually works, and there is no magic in it: → 2,000+ electrodes sit over the part of his motor cortex that runs his hand and arm → those neurons fire differently when he intends to move up versus right. intends. before the arm gets there → they let him play a few minutes with a real joystick while streaming the firing rates out wirelessly → a decoder learns the mapping between the pattern and the movement → then they unplug the joystick and here is the part nobody talks about he kept moving it anyway out of habit. hands on a dead controller, paddle already responding to the signal in his head, and he never noticed the stick stopped mattering that clip is from 2021. five years later you are doing the same thing with GPT-6 Astra the rule files that tell it to read three docs before every edit. the "always run the tests" line. the step by step recipe written so it does the task exactly how you would have. all of it is a joystick you are still gripping, wired to a model that already reads the intent behind it OpenAI wrote this down in their own developer guide. most of what people keep in their setup is compensation for a model that no longer exists, and the compensation is now the defect Pager needed two thousand electrodes and brain surgery to stop using the joystick