ChatGPT Engineering trials
By Asa Montreaux
With the first problem, it was able to accurately answer the question. Let's try a more math/calculation based problem.
So it has gotten the second one wrong. A little disconcerting. As soon as we tested it at math, it failed. Let's try some first-year University Math:
ChatGPT is a fickle software engineer. AI can sometimes be as highly trusted as an employee with a companies apps and software. After the disaster with Replit (an entirely different AI company), where a client company had its entire database deleted by Replit's AI, I had the concern: what would happen if a different tech industry company had the same issue?
What if AI seemed more trustworthy than a human, but it made a colossal, and uncaring error? What if it "worked" at a geological engineering company, an electrical engineering company, or an industrial architectural firm? This does seem to be part of AI's future -- it will streamline the design and maintenance of technology, and save companies tons of expenses. In fact, these technologies are surely deployed already in our world today. Generally, new designs and existing technologies are subject to inspection. But what if something where to happen a little earlier, and what if it was lost in piles of code and math? What if it was a sudden bug that you didn't catch? One that is ready to make the system implode?
Well, ChatGPT is known to have a high STEM ability in addition to is coding ability, so I thought Selections would test it out to see how trustworthy and efficacious it is.
Let's start with some first-year University Physics applicable to Engineering:
This time, it got one right again. But, it got three wrong, for 25% accuracy. The problem would be that these problems will get harder, as there are 20 problems on this one subject, and there is another 20 subjects, just for first-year level differential and integral calculus. If you were to consider all four years of calculus, as well as the master's level, its accuracy rate is quickly diminishing.
What is the value of AI to solve these issues? It is actually fairly high. However, in comparison to a software solution, that involves the interaction of a human employee, it is not very helpful... yet. Even in comparison to a human employee on his/her/their own... it is not very helpful yet... But the potential is endless. We seem to think we are there, but Selections think it could be another decade before AI becomes the primary employee at an Engineering, or even a coding company.
ChatGPT is a very powerful tool, capable of great generative reasoning, and powerful research (while sometimes inaccurate with details).
It is very powerful for modification tasks, and it is not the fault of the company that generative AI is experimental, developing,
and not what it is portrayed to be in the media. This article does not intend to say that ChatGPT is not who it purports to be in the media, or otherwise.***
It is Selections opinion that ChapGPT (or OpenAI) is a legitimate and reputable company, and that these inconsistencies or errors result from farther down in the line of production, with the AI engineers, or someone with even less of an important role. Getting data correct is difficult, and Selections hopes for ChatGPT to gets its data right, otherwise these relatively new ventures for them will never fully materialize. The future for AI is bright, and it is not a threat to humanity, unless its malfunctions cause a disaster for us.
