ChatGPT: Optimizing Language Models for Dialogue
Isaiah Yap
ChatGPT is all the hype recently, with Microsoft investing a multi-billion dollar sum into the company at the time of this writing. If you have been living under a rock, ChatGPT is a chatbot that is powered by OpenAI. It was launched in November 2022 and has been touted as one of the best AI models out there.
The team at Stoic has been testing out ChatGPT’s various capabilities in order to see if the AI model is really as good as it has been marketed. We went through a variety of different research topics, from more complicated tasks such as explaining various medtech, biology, and chemical science terms, to the maritime industry, and to simpler and easier tasks like how to talk like a Singaporean and planning a trip in Europe!
Without further ado, let’s begin!
Complex Information
The first test that I gave to ChatGPT was to ask a basic request. I decided to give him a name, and I called him Jarvis, you know, so I can feel like Iron Man. From here on out we’ll call him Jarvis since he can remember his own name anyway (and I look like Iron Man).
I rate this a success! Next, I used Jarvis to assist me in my day job (as an intern, but I do the actual legwork on most days of course) of researching various ideas and companies that come from multiple different industries. The first task I gave him was to help me understand what nano-glass was!
Although the information seemed superficial, it did give a quick 30-second understanding of what nano-glass was all about and its applications. A 7/10 I would say.
The next test I gave him was on doping of graphene. Here are the results;
Definitely a pass for Jarvis. Not only did he explain what was graphene and how it was doped, but it also gave me the 2 main ways for how it was doped and why those were important. Bonus points given this was a more complicated topic compared to nano-glass.
This will then be Jarvis’s next test! I decided to see if he was able to summarise and simplify it in a way that a layman with no experience (like me), would be able to understand. He was able to cut down the size and length of the article in half, looking like this.
I decided to further test Jarvis on writing super concise paragraphs on the same topic. The result of a 4-cycle summary was this:
No mention of doping, super simple but with some of risk of content being lost in translation. Still a pass, I doubt I would have been able to summarise it to this extent in 4 attempts.
Basic Requests
How would Jarvis deal with the mundane? As I fret over my imminent loss of freedom in serving my nation, I was eager to plan a trip to Europe (ex-Ukraine and Russia). Why not have Jarvis do the planning for me? Starting off with a 3-day trip to Milan, I told him I wanted something a little bit more thrilling, and this was the response:
An iffy pass? Maybe my impression of Italy has been tainted by the old classic trilogy of The Godfather. A boat tour around towns didn’t sound very thrilling 😛
However, when I split the questions, Jarvis fared better, suggesting that I go skydiving, escape rooms (think Hostel), go-karting, quad biking etc, which did seem thrilling.
Back to work. Let’s see if Jarvis can help me with my legal research. We wanted to learn more about Singapore’s Moneylender Act including the associated regulations on interest rates and fees. Here are the results.
Well done Jarvis! I did this after I had handed in my research much earlier, of course, no help from Jarvis on that. Jarvis was able to quote legal documents issued by various authorities to give the various clauses and sections to help with further research that the user might be more interested in.
ChatGPT’s output may not always be perfect or accurate. One example was when the bot was taking an MBA exam set by a Wharton professor. The bot made mistakes in simple calculations of 6th-grade math and could not handle advanced processing analysis questions. Maybe with GPT4 Jarvis can function and perform more independently, subject to some supervision and fact-checking. But where would that leave us simple humans?
Disclaimer
Please refer to our terms and conditions for the full disclaimer for Stoic Capital Pte Limited (“Stoic Capital”). No part of this article can be reproduced, redistributed, in any form, whether in whole or part for any purpose without the prior consent of Stoic Capital. The views expressed here reflect the personal views of the staff of Stoic Capital. This article is published strictly for general information and consumption only and not to be regarded as research nor does it constitute an offer, an invitation to offer, a solicitation or a recommendation, financial and/or investment advice of any nature whatsoever by Stoic Capital. Whilst Stoic Capital has taken care to ensure that the information contained therein is complete and accurate, this article is provided on an “as is” basis and using Stoic Capital’s own rates, calculations and methodology. No warranty is given and no liability is accepted by Stoic Capital, its directors and officers for any loss arising directly or indirectly as a result of your acting or relying on any information in this update. This publication is not directed to, or intended for distribution to or use by, any person or entity who is a citizen or resident of or located in any locality, state, country or other jurisdiction where such distribution, publication, availability or use would be contrary to law or regulation.
