1 Step Forward, 2 Steps Backward - AI Journey Thus Far

Since I am in-between jobs, I get more time to sit & ponder over a number of things like my finances for the coming months and what to cook for dinner. However, what really got to me were the 3 rejections I faced from companies where I had applied for a job. A kind friend offered solace & help by sharing links for AI resume builders online and job search app which applies to places on your behalf.

And that got me thinking how LAZY am I? How much help do I need? Do I really need AI to tell me where I shine and where I s**k??

Read below to find out what I researched and found out about AI....

Trust me, You NEED TO KNOW this!

Fact I: AI data centers consume a significant amount of energy as the process requires access to high-performance computing resources, including powerful CPUs, GPUs, or specialized AI accelerators like TPUs.

As per research paper, training GPT-3 model required the equivalent of 405 years’ worth of energy consumption from Nvidia V100 GPUs, which are known to consume around 300W of power which amounts to appx 1100MWh per year. This means that per day energy consumption of ChatGPT-3 = 3MWh and for GPT-4 must be greater.

It has been predicted that by 2027, electricity consumption by AI models will reach 100+ Terawatt hour annually.

Now let's discuss its green house gas emissions.

Fact II: It releases approximately 1.3 tons of Carbon Dioxide per day which is equivalent to the emission from burning 1413 pounds of coal. This equals release of 460 metric tons of CO2 annually which is comparable to those from 58 homes in a year or 1million+ miles driving in a gas/fuel based vehicle.

This next fact will certainly shock you if the others haven't yet.

Fact III: Training of GPT-3 uses lots of water. Since it's being kept all hush! hush!, the range can be anywhere between100-700 thousand or even millions of gallons. On an average, ChatGPT uses 500ml water for processing and conversion of 10-50 prompts & their responses.

It has been calculated that by 2027, freshwater consumption by AI models will be 4.2-6.6 billion cubic meters. This is clearly adding on to the existing water scarcity that 2/3rd of world population faces already.

I recently read that a top electric bicycle company, has added ChatGPT, to its e-bike. This allows riders to have voice interactions with their bike and to obtain a wide range of information and support.

While the concerns for safety and data privacy can't be ignored; the bigger question is who needs it? Do you?

I am not a biker but I know those who take it too seriously for racing while others do it as a fun activity outdoors to connect with nature. As it is our gadgets overpower our social interactions, why allow others to butt in even more into YOUR personal space - YOUR "me" time.

So, what can be done?

Should we stop developing further or boycott its use entirely?

No! That is neither my intent nor a solution. Here are a few suggestions to make note of:

Kind request to developers

Tip 1: To make it more palatable to eco-friendly users as well as the planet & its environment, it has been recommended to NOT create LLMs, unless it is bound to lead to great use and importance.

Tip 2: Instead, improved sparsity regulation, low-precision arithmetic and efficient algorithms that optimize data processing should be used to reduce processing energy, time and emission.

Tip 3: Moreover, the developers need to work towards improving runtime water efficiency as water has been, is, and will remain a finite source and NEEDS to be used judiciously.

Kind request to users

Before typing in your next query on Google or next prompt in Gemini, think for a second and analyze do I really need to run this query? Can I do it myself?

Isn't my school or college assignment supposed to be a learning experience?

Isn't my resume supposed to reflect who I really am?

Isn't there a library to read that book my professor recommended for Advanced Mathematics?

Why not pick up that novel I bought from Book Fair which has been collecting dust ever since?

Won't my grandmother love to tell me about Ramayana?

Isn't my father capable of explaining the difference between mutual funds & SIP?

Do I really need that YouTube video to learn to make pakodas when I can simply go to the kitchen and see her doing the same LIVE every time it rains?

As much as we should be grateful to be living in times of great technological advancement, but shouldn't we be thoughtful of the consequences of our actions too???

The usefulness of any tool, LLM, software or device isn't in question here and nor is it up for debate. I myself use so many of these everyday for work. Hereon, I will choose to actively pause and post my next query, only when needed.

Will you too?

Another fact to wrap up: Google Search also utilizes approximately 25-30TWh of energy yearly.

Next week: Energy & Data Usage per person per day