Any matrix multiplication can be trivially unwrapped into a bunch of for loops with basic arithmetic. In fact, that is exactly what any GPU kernel ultimately does in some sense.
What differentiates a “traditional computer program” from a single matmul? How about a bunch of matmuls? How about a bunch of matmuls with some non-linear arithmetic thrown in between them? And if we are adding in sampling from a distribution but set temperature to zero and use random seeds so that it is fully deterministic?
What you seem to be saying is that the machinary your traditional computer program executes on is capable of executing intelligence.
What I said is thst a traditional computer program (say Excel) is not intelligent in itself, as it cannot, itself, without further input from a human, adapt.
> What I said is thst a traditional computer program (say Excel) is not intelligent in itself, as it cannot, itself, without further input from a human, adapt.
Excel is as capable of adapting to new input as any AI, and the input need not come from a human specifically: the act of adapting to input is an automatic consequence of being able to perform linear regression.
In fact, this is the real value of a spreadsheet over a piece of gridded paper where the sums are done by hand.
If "adapting to input" is your test for intelligence, however, it is a test which a thermostat will pass.
>Excel is as capable of adapting to new input as any AI, and the input need not come from a human specifically: the act of adapting to input is an automatic consequence of being able to perform linear regression.
I said 'sensory input', as in, "any primary, real world data". Can I just download excel, and without feeding it machine learning algorithms, feed it a bunch of videos or text and have it learn history?
>If "adapting to input" is your test for intelligence, however, it is a test which a thermostat will pass.
Not just adapting to specific input the system is designed for, adapt it to ANY input. A thermostats function will never change on its own without modification, just like excels, or whatever other traditional program.
What differentiates a “traditional computer program” from a single matmul? How about a bunch of matmuls? How about a bunch of matmuls with some non-linear arithmetic thrown in between them? And if we are adding in sampling from a distribution but set temperature to zero and use random seeds so that it is fully deterministic?