You are free to try and run LLMs on your old hardware. Don’t be surprised when (not if) it takes 4 business days to generate completely nonsensical replies to the most basic prompts.
One of the KPIs for IBM engineers used to be kLOCs or “1000
lines of code”. This encouraged the engineers to write more complicated, obfuscated, inefficient code.
If you cleaned up your code to be more efficient, you were penalised.
While this doesn’t make sense to developers and users, it does make sense to hardware sales staff. The crappier a program runs on the existing hardware, the more incentivised customers are to buy new hardware.
My argument is that by adding technologies branded as “AI” to software platforms that inefficiently consume system resources, software developers don’t actually add anything to the platform. It is just a way for manufacturers to create more eWaste through forced obsolescence.
You are free to try and run LLMs on your old hardware. Don’t be surprised when (not if) it takes 4 business days to generate completely nonsensical replies to the most basic prompts.
One of the KPIs for IBM engineers used to be kLOCs or “1000 lines of code”. This encouraged the engineers to write more complicated, obfuscated, inefficient code.
If you cleaned up your code to be more efficient, you were penalised.
While this doesn’t make sense to developers and users, it does make sense to hardware sales staff. The crappier a program runs on the existing hardware, the more incentivised customers are to buy new hardware.
My argument is that by adding technologies branded as “AI” to software platforms that inefficiently consume system resources, software developers don’t actually add anything to the platform. It is just a way for manufacturers to create more eWaste through forced obsolescence.