Dear
Friends
Caution:
I am writing this article in the context of Public Awareness and Futuristic
Opportunities. All views are personal.
This
morning I received a Video of Dr Matt Welsh, Co-founder of Fixie.ai, a
Seattle-based startup developing a new computational platform with AI at the
core, Ex-Apple, Xnor.ai, Google, and a Professor of Computer Science at Harvard
University, PhD from UC Berkeley.
This
has been forwarded by my former colleague & friend (Mr. Jerry Kurian) at
Quidnunc Development Centre, and currently, Founder @ "GenAI People"
- Mentoring Program.
I
watched the 1:06:55 hr video (Large Language Models and The End of Programming
- CS50 Tech Talk) in a single go!
(www.youtube.com/watch?v=JhCl-GeT4jw).
Let’s
talk about Numbers first!!
Based
on the US Salary, the 1-day cost for a regular experienced software developer
including all insurance/perks/total cost to the company is $1200. Assume on
average, the number of lines of code checked into the system (final tested and
useful) is 100 per day per developer.
Average
number of GPT-3 tokens per line is 10. The price for GPT-3 is $0.02/1000
Tokens. Hence total cost per developer per day by GPT is $0.12 against
the developer cost per day is $1200.
If we
take Indian cost, assume the Total cost to the company per developer is Rs
30,00,000. Per day cost is Rs 8300!! Whereas GPT equivalent cost is less
than Rs 10!
Why
does an entrepreneur hire more developers when technology brings down the cost
of conventional development barring project management, requirements specs,
customer interaction, etc?
What
are the additional roles companies would invest in?
1. Project
Manager who manages the team
2. Product
Manager who understands the domain/product
3. Test
Designer cum Prompt Engineer who tests whether it works or not and also changes
the code using alternative prompts if it fails in the test!!
What
Qualifications are needed for the above roles?
- Common
Sense, Critical & Design Thinking, Domain Understanding
What
happens to the existing Computer Science Course Structure in the colleges?
ü
I have a counter question for you. Are we really worried
about how Low-level machine-level things work (processor/memory level)
after the introduction of high-level programming languages esp. Python? Of
course, it is useful for certain applications like OS/IoT, etc. I am talking in
a large context. Down the line, the same thing might happen to Python. Python
might become a Low-level language when LLM develops applications and English
Prompt becomes a high-level language for us!
ü
If LLM does the entire computation, stores
it in its Vector databases, capable of optimizing the algorithm, why should
programmers worry about
o
Which Data Structure should be used?
o
How do we store and retrieve from DBMS using SQL query?
o
How do we deploy, where to deploy the code, how fast etc
(DevOps)? It is the botheration of LLM but not the developers! It does it in
seconds!!
o
In such a scenario, Do I need to learn Computer
Networks/OS/Cloud Computing/Software Engineering? Will they become obsolete by
that time?
ü
What happens to existing degree holders and
future degree holders?
o
Perform with the Prompt, else, Perish with your Procrastination
o
Campus selection happens but based on your ideas, domain
understanding, prompt crafting skills
o
There won’t be the Dept of CSE alone! CSE gets merged with every
Department and Domain!!
ü
The role of non-CSE with an example?
o
Let us consider Civil Engineering
o Engineers might use natural language prompts to describe
project requirements, and the LLM could
generate initial design sketches or suggest improvements based on its vast
knowledge of engineering principles.
o
Engineers could describe project specifications, and the LLM
could produce code snippets for simulations, analysis, or modeling.
o
Detailed documentation and reports by LLM
o
Large Language Models might interface with Building Information
Modelling software, allowing engineers to communicate with the system using
natural language.
o Risk Analysis and Decision Support by LLM with Civil Engineering
Capabilities. I am not joking!! Very soon it is possible!!
ü
You know, KissanAI unveiled Dhenu 1.0 LLM for India’s
Agricultural context!! The
model's uniqueness lies in its bilingual nature, processing 300,000 instruction
sets in English, Hindi, and Hinglish.
ü
Krutrim, ‘India’s own AI' developed by Ola
CEO Bhavish Aggarwal/Krishnamurthy Venugopala Tenneti-led
venture (krutrim si designs)- trained
2 trillion+ ‘tokens/ sub-words used in conversations and capable in 20
Indian languages. larger
than even GPT-4 in Indic language support
ü The day is not far. Have you seen Amazon’s fulfillment centers?
How do robotics,
miles of conveyors, complex scanning & sortation equipment, advanced automation, and
machine learning come together with the help of AWS to get packages to
customers in as fast as one day for Prime members? Future Factories are
built in this way!
Are you Convinced? Regardless, I am!
Raise
your Bar! Otherwise, Get Rinsed in this Wave!!
Dr.Ravi
Saripalle
No comments:
Post a Comment