Running AI consumes a lot of energy, make sure you are using it in wise applications.
Last 3 three days I have been experimenting with some offline AI models, especially the vision models (diffusion models, clip models, lora models, etc) and it made me run my human thinking:
Are we humans in right direction to use such exponentially insane energy monsters?
MATH doesn’t fits in for me from sustainability point of view. RAM and GPU goes to almost 90% even with an 8GB model. CPU temperature beyond 50deg celsius – Just for simple use cases e.g. summarising two paragraphs, asking a summarising text, asking a question which can be easily googled, etc. basic tasks which are very easy for humans.
IT’S NOT EFFICIENT.
As my readers, please KNOW THAT -> AI never THINKS. A MACHINE can never be CONSCIOUS.
In reality,
It’s does PATTERN MATCHING
WORDS
to numbers
to massive calculations
to numbers
back to words or pixels
only when a HUMAN ASKS with their CONSCIOUSNESS.
Most models are based on set of data which are not REALTIME / COMPLETE in its true sense. Think of AI MODELS as huge files which need to be re-generated for including the newest data-sets which humanity creates. Call it Training or Fine Tuning, in this context it’s the same thing for me. More re-generations for making it light weight, at cost of quality.
Comparatively, HUMANS are more flexible and smart in many contexts as compared to AI models alone.
CONFIDENCE for example is just a number for AI models. I am writing this blog because I want CONSUMERS of AI to be FULLY aware about it, only then you can make the WISE decisions out it.
It’s NOT a bad TECHNOLOGY, it’s the APPLICATION we make out of it should be NET EFFICIENT in terms of experience, cost, energy and clicks.
This is extended version of my previous AI decision making framework.
IF WE ARE giving so much attention to GPUs
I really don’t understand WHY we would not NOT give
“ATTENTION” to
HUMANS and HUMAN health worldwide? They are DATA CREATORS of those MODELS!
It’s a real tough question to ask ourselves.
A good thing about AI is that it can be implemented in REAL GOOD USE CASES.
In a way which can even REDUCE the net energy consumption. You just have to figure out the smart areas where you can apply these AI processes with right mindset.
Even if you are generating images/contents using AI, make sure it helps add value to viewers life in a positive way and which are real hard to generate in other setups.
It’s a time vs. value vs. energy race.
HERE, A few images created during this human thinking:
