Oh definitely, don’t get me wrong: I like this platform all in all. It’s fun to operate, especially when it’s completely new. The older the model the more finicky it gets.
I gotta say the CPU is my favorite feature, though. I just can’t get over the fact that I can power a neural network with multiple specialized subsystems and a frankly impressive language processing accelerator with a literal bowl of oatmeal. Try putting oatmeal in your NVIDIA NN accelerator and see how that goes, but give this fucking thing some oatmeal or a bit of cheese or whatever and it can process input from the actuators, the visual and audio sensors, pressure-sensitive sensors etc all simultaneously while it’s doing language tasks and eg. navigating, and still respond to surprising changes in the environment in just a few hundred milliseconds at best, and some of the models can even be used to generate things like art or term papers or weapons systems designs. They do have a tendency to give you bullshit answers, so you have to be careful not to just blindly trust their responses.
The CPU is also my least favorite feature though: while it’s an amazingly cool design considering how fucking old it is –I think some of the prototypes for the firmware that regulates the generator and oxygen pumps etc. are from over 500 MYA which is fucking nuts – it really wasn’t designed to be networked with hundreds of millions other units and that can cause no end of mysterious bugs and issues. Unfortunately that’s the way everybody’s setting them up nowadays, and the external mobile network devices that are now ubiquitous are making the problem much worse and causing even weirder problems. Like, who would have thought that keeping the unit connected to a network device for too long or connecting too often can actually trigger bugs in the CPU’s firmware, because its heuristics were tuned to communicate with and identify no more than a few hundred units and they start giving wrong answers that get worse as the network grows. Run the unit in an overcrowded network for too long too often and eventually even its performance will drop, and at worst something like the threat detection system can suddenly shit the bed and start doing stupid shit like rejecting the credentials of every unit outside of its current subnet. Sometimes this bug can actually spread inside a subnet, and cause it to start interfering with neighboring or even fairly distant subnets, just spawning logic bombs left and right.
Oh definitely, don’t get me wrong: I like this platform all in all. It’s fun to operate, especially when it’s completely new. The older the model the more finicky it gets.
I gotta say the CPU is my favorite feature, though. I just can’t get over the fact that I can power a neural network with multiple specialized subsystems and a frankly impressive language processing accelerator with a literal bowl of oatmeal. Try putting oatmeal in your NVIDIA NN accelerator and see how that goes, but give this fucking thing some oatmeal or a bit of cheese or whatever and it can process input from the actuators, the visual and audio sensors, pressure-sensitive sensors etc all simultaneously while it’s doing language tasks and eg. navigating, and still respond to surprising changes in the environment in just a few hundred milliseconds at best, and some of the models can even be used to generate things like art or term papers or weapons systems designs. They do have a tendency to give you bullshit answers, so you have to be careful not to just blindly trust their responses.
The CPU is also my least favorite feature though: while it’s an amazingly cool design considering how fucking old it is –I think some of the prototypes for the firmware that regulates the generator and oxygen pumps etc. are from over 500 MYA which is fucking nuts – it really wasn’t designed to be networked with hundreds of millions other units and that can cause no end of mysterious bugs and issues. Unfortunately that’s the way everybody’s setting them up nowadays, and the external mobile network devices that are now ubiquitous are making the problem much worse and causing even weirder problems. Like, who would have thought that keeping the unit connected to a network device for too long or connecting too often can actually trigger bugs in the CPU’s firmware, because its heuristics were tuned to communicate with and identify no more than a few hundred units and they start giving wrong answers that get worse as the network grows. Run the unit in an overcrowded network for too long too often and eventually even its performance will drop, and at worst something like the threat detection system can suddenly shit the bed and start doing stupid shit like rejecting the credentials of every unit outside of its current subnet. Sometimes this bug can actually spread inside a subnet, and cause it to start interfering with neighboring or even fairly distant subnets, just spawning logic bombs left and right.