Learning about "Will ChatGPT take away all jobs?" - My "Chamka" moments week of 1/27/2023

Ever since ChatGPT made its appearance on the global stage, it's been the topic of conversation (no pun intended) across social media posts, news articles, friendly discussions, work meetings, schools…. My wife and I have even been discussing it in our dog-walking time together. I've spent more time in the last 3 weeks than I should have on prompt engineering, writing python code against OpenAI APIs, reading and discussing the implications of what is claimed to be a cataclysmic upheaval. My Chamka moments this week come from these experiences.

1. Leveling the playing field -

Like many others, my interactions with ChatGPT started with poem generation, code snippet debugging, research summarization. I was surprised at the ability of ChatGPT to not just understand the context and intent of the conversations, but to synthesize answers that cut across its vast training data, with the added human elements of creativity, emotion, audience awareness (try having ChatGPT write a poem with the same context, but varying in sadness/happiness, for an elementary school student vs an elderly couple).

Impressed by these initial interactions, I wanted to test ChatGPTs abilities to take on more complex problems like text analytics, clustering random sentences, classifying sentences into predefined clusters. Of course, I had to "program" (aka prompt engineer) ChatGPT to understand my inputs, provide outputs in consistent ways. Given the repetitive need for prompt engineering, I decided to write python code to call OpenAI's underlying GPT-3 APIs. Not having coded for years, I found myself going to Google for basic python commands. If lucky, the instant answer from Google displayed what I needed on the search results page, without having to click through other links. Most of the times though, I had to click into multiple links, ask my question in different ways, to get an answer. It was a slow process as I fumbled my way through re-learning the syntax of python commands. The next morning, a glaringly obvious thought came to mind - why don't I ask ChatGPT to generate code to emulate ChatGPT using OpenAI APIs. Sure enough, ChatGPT gave me some well documented, simple code to call the text-davinci-003 completion endpoint. It didn't include code to carry context between conversations, but thanks to this article by Daniel Abrahamberg,?I was able to create and pass this context. A few questions to ChatGPT later, I had written "SidGPT" - the following simple code that, when run in an IDE - like VSCode - ran a close equivalent to ChatGPT right in context of where I was writing code. Which turned out to be quite a blessing, since ChatGPT itself became quite flaky over the following days. With SidGPT by my side, I was able to rapidly get answers to questions like "how do I write python code to add items to a hashtable with a string as a key and a list of strings as values?", copy/paste code from SidGPT into the IDE, and with a little prompt engineering, build a pretty decent text classifier.

For someone who hasn't coded in years, but has lots of ideas, this experience was a boon. OpenAI's APIs have leveled the playing field for me, so that I don't need someone else to prototype my ideas - with SidGPT by my side, I can code well enough to build basic proofs of concept - if I can articulate an idea clearly enough, I can code it. I will never take the place of expert coders, but I will not need as much help prototyping. And this was a "Chamka" moment - AI is on a path to leveling the playing field in many ways : non-coders writing code, non-poets writing poetry, non-native language writers writing fluently, non-search engine companies building specialized search engines. As I learnt from my experience between Google searches and ChatGPT prompts, I no longer like the old search paradigm of "scour the web to find me snippets of what I'm looking for, and let me stitch it together". I prefer - "let me tell you what I want to do, you stitch a personalized set of steps for me", or even better "do it for me". While Google has had LaMDA tech for many years, I believe that OpenAI has democratized Large Language Models in a way that the battle for search dominance could take a brand-new turn (an evolution from search engines to "personalized task completion engines"?).?In what other ways do you believe AI can help level the playing field?

Here's SidGPT.py, a convenient script that emulates ChatGPT and runs within your favorite Python IDE, as you code your next amazing project -

import o
import openai
import csv
import sys

openai.api_key = "<add your API key here>"

# Keep collecting context from prior prompts and responses
context = ""

# Keep the prompt open forever
for line in sys.stdin:
??if 'q' == line.rstrip():
????break

??context = context+line
??# trim context if it is getting too long, if we don't the completion API will reject it
??if (len(context) > 7700):
??   reductionneeded = len(context) - 7700
??   context = context[reductionneeded:]

??response = openai.Completion.create(engine="text-davinci-003", prompt=context, max_tokens=2048, temperature=0)
??print(response["choices"][0]["text"]+"\n")
??context = context + "\n" + response["choices"][0]["text"] + "\n"        

2. Moving up Maslow's pyramid -

Among the many conversations that ChatGPT has sprung up, one is - will automation take over our jobs? Clearly, no one has a magic crystal ball, and answers vary from Never to not in the near future to Yes, it's impending for certain jobs. However, if we turn to the past as a barometer for the future, we can draw parallels to other disruptions - huge ones (Industrial Revolution, Computerization, Outsourcing), and others (Ubiquitous mobility, Uber-ification, Assembly programming to OOP, Calculators instead of mental math). Each of these disruptions changed the way we lived our lives, education, work. BPOs took over data entry jobs, Machines took over farmer jobs, computers took over typist jobs, Uber took over taxi driver jobs. Parents worried about calculators rendering their kids mathematically incapable, and mobile phones obliterated land lines. Every disruption shook the status quo, created value we didn't think possible before, created space for society to climb up the proverbial Maslow pyramid, rewarded those who did climb up, and left behind those who didn't.

In a recent conversation with Sachin Singh , Rish Tandon , Abhoy Bhaktwatsalam , we were discussing how close to the top of the pyramid we had already reached. Are we so close to the top that there isn’t much left to climb? Is this the time when machines will finally take over our jobs? Who knows - but one thing is for sure, by making research, coding, learning easier, this disruption is bound to make us more efficient. And increased efficiency means a reduced need for humans …. unless those humans have reskilled themselves to do what the new technology cannot do. Farmers learnt to drive tractors, typewriting secretaries had to educate themselves on computers, taxi drivers had to skill themselves on mobile phone applications to drive for Uber. The AGI wave unleashed by ChatGPT appears to me like another big revolution - each of us should be looking at ways to upskill ourselves to leverage AGI and climb Maslow's pyramid.

Like most technology disruptions, I would expect this disruption to hit the trough of disillusionment at some point soon, especially given how inflated our expectations are -

3. Is creativity the human super power that places us higher on the pyramid than AGI? -

It was quite enlightening to listen to Sam Altman in this youtube video where he describes how the traditional thinking of where AI would benefit society most, was proven wrong. The traditional thinking is that AI would first replace manual, blue collar jobs, then advance to repeatable white-collar jobs, then knowledge workers, then research, and at some point, it would bring in creativity - which is a fundamental human super power. Sam argues that AGI has reversed the traditional thinking - ChatGPT is able to do creative tasks better than many humans.

In the book Where good ideas come from, author Steven Johnson calls out how creative geniuses like Edison relied on reading a lot, networking, and allowing new ideas to form through cross-linking of existing ideas and thoughts. How different is ChatGPT? It has been trained on 50TB of data, 1.5 billion conversation, it's models apply learnings across domains to build new connections. When ChatGPT generates poetry, parodies, research summaries based on a supplied audience IQ, that's creativity. When Dall-E generates Van Gogh style renditions of imaginary scenes, that's creativity. Sure, AI gets it wrong many times, but if we use it to bootstrap our thinking (not as the final output), it's a huge creativity boost. According to Forbes, ChatGPT will enhance human creativity, not destroy it. So, if AGI is helping us get more creative, and is capable of learning how to get more creative itself, the question comes back to - what is our superpower that will keep us a step ahead of AGI? Is it "Content Creation", "Problem framing", "Knowing what questions to ask?", "Super specialization in domains",?"Fact checking AGI", "Post processing AGI responses", something else? What is your superpower that you believe will give you an edge over AGI?

?

It's a time of great AI advancement. Multimodal learning, reinforcement learning emulate how babies learn. And if you've been around a baby, you'll know how they learn in surprising (and creative) ways. Is the tidal wave that will sweep up all jobs here? I personally don't think it's landed yet, but I do believe it is building up. It will go through the peaks and valleys in the Gartner hype cycle, but IMHO, this wave will hit the shore (unlike crypto) and will cause us to move uphill. It's not a question of if, it's when. How are you moving up the value chain?