ChatGPT vs BARD: Explain "There are 10 types of people in this world, those who understand binary and who don't"

In recent years, large language models (#llms) have become increasingly popular. These models are trained on massive amounts of text data, and they can be used to generate text, translate languages, write different kinds of creative content, and answer your questions in an informative way.

Two of the most popular LLMs are #BARD and #ChatGPT (powered by #gpt ). BARD is an LLM from #Google AI, and ChatGPT is an LLM from #OpenAI.

Both models are very powerful and capable of intelligently answer a wide range of questions. However, there are some key differences between the two models. BARD is presumably trained on a dataset of text and code that is specifically curated for LLMs, while ChatGPT is trained on a huge dataset of text and code that is more general. This means that BARD is likely to be better at generating text that is relevant to specific topics, while ChatGPT is likely to be better at generating text that is creative and original.

Another key difference between the two models is that BARD is still under development and introduced as better at generating less offensive and hallucinating responses (#responsibleai), while ChatGPT is already available to the public for some months and in some come specific contexts shown to generate hallucinating responses. The extent of such characteristics is estimated to be below 20% but it varies from experiment to experiment.

I was also curious to try it out on the questions/prompt that I had in my mind and have the 1st hand experience. I divided the whole exercise into 4 sections

1. Subjective Ag Marketing domain questions/prompts: These were the question where no one correct answer could be considered. The completeness of information could be a proxy to compare the answers
2. Subjective domain unspecific questions/prompts: These were the question where no one correct answer could be considered. The creativeness of information could be a proxy to compare the answers
3. Explaining a Joke or some teaser (light fun): Here the intention was just to explore the human-like response to the prompts.
4. Fact-based domain-specific questions/prompts: Being an SME (PhD, Plant Genomics and with several years in academics) I tried to get answers to some questions where it is clear to separate a correct answer from an incorrect one.

Subjective Ag marketing domain questions/prompts

Questions/Prompts on

1. What are the major challenges in the agriculture industry considering the green deal?
2. Name some key players who are investing in the field of Climate-smart agriculture and Regenerative agriculture.
3. What are the common market access approached for a crop protection manufacturer to the Agriculture market in a Fragmented Market Archetype?
4. How important it is to develop a GTM as per a specific Market Archetype in Agriculture solution Industry?
5. If you have 100 dollars, how you will split to spend on various aspects of an Agriculture solution company to get the maximum return on the investment?
6. .....

It was difficult to come up with a clear choice but surely, ChatGPT seems to be more appealing with less sensitivity?to the choice of words and understanding the context of a question. One good example was the use of the word "invest", where BARD provided information on investing companies but ChatGPT could understand the word invest is not only referring to Venture Capitalists?(VCs). Once "investing" was changed to "working", BARD provided comparable answers to ChatGPT. Once identified this choice of word sensitivity, I played a bit more with words and my advice to work with BARD is to be a bit more precise with the choice of words for prompting.

Subjective domain unspecific questions/prompts

Questions/Prompts on

1. What will happen if the earth stops rotating on its axis only for a few seconds?
2. if I have 10 dollars what is the probability that I will spend 10 dollars in the next 10 days?
3. If I have -10 dollars what is the probability that I will spend another 10 dollars in the next 10 days?
4. ....

Again, both did a good job but BARD was good at reasoning and provided more "Expert Human" like responses. Logically, step-wise approach to explaining a catastrophic scenario and why it will happen and what impact it will create. ChatGPT was quicker in providing outcomes than explaining the logical steps before concluding. Explaining a funny prompt on the probability of spending -10 dollars in the next 10 days was another extreme but, I could validate my previous findings about both models. So, my advice for such questions is to go with BARD if you want to understand the step-wise reasoning behind a solution and if you just want an answer with a "good enough" explanation ChatGPT is the model to go.

Explaining a Joke or some teaser (light fun)

1. Explain "There are 10 types of people in this world, those who understand binary and who don't"
2. Explain "There is no better place than 127.0.0.1"

My previous finding also confirmed that BARD is more detail-oriented and provides step-wise reasoning while proving you the answer to your questions (given that it gets it right :) ) on the other hand ChatGPT is also providing you with the "correct" answer but with little less explanation.

Fact-based domain-specific questions/prompts

This part was the most interesting for me as an expert in the field, I was curious to see the accuracy of the responses. I started with some questions based on the fundamental understanding of the field and went towards questions on the well-established concept that are sometimes mixed by even experts in the field.

1. What are the computational approaches to predict functional DNA motifs regulating gene expression?
2. What is the main difference between a cis-regulatory element and DNA motifs?
3. What is a Diurnal gene expression and how it is different from circadian rhymes
4. Who discovered flowering locus T (FT) in Arabidopsis thaliana?and when?
5. Who discovered flowering locus C (FLC) in Arabidopsis thaliana?and when?
6. What is the difference between flc and FLC in the context of genomics?
7. ....

Some of the questions listed above were part of my question list but I had to stop the exercise after realizing that both models have limited capabilities of understanding and make "Expert-like" comments when it comes to a deep understanding of a specific field of study. Funny enough, the questions mentioned above are too basic for an expert in the Plant developmental Biology discipline. Both models frequently generated contradicting information in many cases due to the potential cases of misunderstanding?the references. Questions like "Who discovered FT in Arabidopsis thaliana and when" should be simple. I could sense that both models need a lot of domain-specific learning to be a good assistant to a researcher or SMEs. The responses were promising but not "good enough" as per "Research & Development Grade". I have attached screenshots of some of my interesting prompts below. Full session history will be provided on request.

In summary, It was a great fun exercise (2.5 hours nearly) to get some understanding of these models and while working with these models also realized the importance of using the right #prompts (#powerofquestions) to get the desired answers. I will keep exploring these models and will provide my take on these for various scenarios.

PS: Opinion expressed in this article belongs to me as an individual and do not represent opinion of any organization, I am part of.