Hands-on review: ChatGPT
Pritish Sehzpaul
Fullstack Software Engineer | ex-Head of Technology at Ikarus 3D | ex-Associate at JPMorgan Chase
This particular bot has caught the attention of people at mass - both techies and non techies.
ChatGPT has been a conversation topic recently. Its context aware and conversational design provides, to an extent, the informal touch that we have always wanted in AI. One thing to note is that no matter how intelligent the bot sounds, the team has taken painstaking steps to ensure that the general public cannot confuse the model with a human. Every time you try to have a conversation that involves having widespread belief, the bot prompts that it cannot have an opinion along with a few sentences of clarification - that ChatGPT is AI trained on a limited dataset.?
There is one difference that is outrightly felt - Jarvis was spot on with its sarcasm giving it a human touch. ChatGPT is designed so as not to have personal opinions or belief systems, leading to its lack thereof.?
Brimming with curiosity ourselves, my team and I took a few hours to chat with ChatGPT. Here’s what we learned.?
What we found from our conversation with ChatGPT?
We asked ChatGPT some questions around the 3D industry and received some unique insights about the sector as well as about ChatGPT’s utility.?
What we could gather is that ChatGPT is best used for general conversations, making it a good fit for conversational AI and chatbots, standard article creation or as an ideation partner. You can conceive new ideas by chatting about your topics with ChatGPT, and build upon the conversations you have with it.
When we asked questions regarding the 3D industry, it was surprising to see that some of the answers were similar to what we have been trying to solve. If ChatGPT can be considered as a collective of data it is trained on, this means that the entire 3D industry is facing the common issue of ownership of 3D assets.?
We also know how custom 3D models can be expensive. A model can range from $20 to $1000 depending on multiple factors. We wanted to know if ChatGPT understood this.
It sure did.
ChatGPT can give answers which seem correct but aren’t. Some of these situations are where we require expert knowledge and skills like balancing a redox reaction, details of a research paper, etc. It can give you the summary or act as a guide of how to proceed, but when it comes to solutions, it can be ambiguous or completely incorrect.
Though ChatGPT’s replies are much lengthier than the answers on search engine results pages, their correctness isn’t as consistent in comparison. In one instance, we asked questions on balancing redox reactions where it produced wrong results, and then asked questions on simple integrations of linear and quadratic functions, which gave us correct results.
Conclusion
ChatGPT is an experiment that gives us a glimpse of what the future beholds. It need not be perfect to exhibit what AI tech is capable of. The ChatGPT team at OpenAI is a brilliant group of people aware of the limitations and have openly published them on their blog - https://openai.com/blog/chatgpt/ . This shows a willingness to improve and the awareness of their current approach.?
Where ChatGPT shows the potential of large language models and their pitfalls, its indiscriminate use can lead to an increase in unchecked information that has no sources, resulting in the spread of fake news.?
With regards to industry specific insights, it is being observed that ChatGPT doesn’t have the potential to disrupt the 3D industry yet other than the fact that it can help ideate over what kind of 3D models the artists can create next. It can take over their mental workload for creation. But what is art without inspiration? A model like ChatGPT that cannot have opinions, will not have inspiration.?
Could CHATGPT be a precursor to Jarvis? Maybe.?
All in all, we found two sides to the ChatGPT experience. One, that it is a fun toy to play with. Second, that it is a promise of greater AI-powered innovations in the future.