There’s something up with the eighth bit (index [7]) of every 64-bit chunk. It has a remarkably low turnover rate (23%) when compared to its next-door neighbors. Bits [48:51] also have low turnover rates (22%-25%), but the eighth bit’s low turnover rate uniquely persists when extended to context windows with lengths up to 20 chunks.
The last seven bits of every 64-bit chunk actually carry only one bit of information. The bit right before these seven (index [56]) has an abnormally high turnover rate w.r.t. next-door neighbors(66%).
Part of me wants to attribute this to some cellular automaton rule. But isn’t it interesting that, in a chunk, the eighth bit is unusually stable, while the eighth-last bit is unusually volatile? Some weird kind of symmetry at play here.
There’s something up with the eighth bit (index [7]) of every 64-bit chunk. It has a remarkably low turnover rate (23%) when compared to its next-door neighbors. Bits [48:51] also have low turnover rates (22%-25%), but the eighth bit’s low turnover rate uniquely persists when extended to context windows with lengths up to 20 chunks.
The last seven bits of every 64-bit chunk actually carry only one bit of information. The bit right before these seven (index [56]) has an abnormally high turnover rate w.r.t. next-door neighbors(66%).
Part of me wants to attribute this to some cellular automaton rule. But isn’t it interesting that, in a chunk, the eighth bit is unusually stable, while the eighth-last bit is unusually volatile? Some weird kind of symmetry at play here.