A few weeks ago I posed a question to ChatGPT. Asking it to size and select the appropriate machinery for a largish but simple cold room design. One that would not cost more than around £200,000.

It took me around twenty minutes and three attempts (adjustments) to present ChatGPT with the right questions. And this don't forget by me, an engineer who has 'been in the business' for more than half a century!

ChatGPT returned its answer in around 10 minutes. Containing three fundamental errors making its design technically non viable. Also. No 'Presentation quality' printed report. Just a stream of largely irrelevant waffle, interspersed with some facts. Jargon ridden and muddled. No recoverable and reproducible digital calculation file. No 'Audit trail' and absolutely no references, citations or corroboration. So 30 minutes spent in total.

For the question I presented, Mistral software would have provided all that ChatGPT failed to provide. Along with a totally accurate answer and it would have done so in under sixty seconds from starting the Mistral program. Furthermore, barely requiring the user to supply answers to any raw questions at all. In other words, mostly multiple choice menu selections via a strong, dynamically error trapped, Graphical User Interface. Meaning no knowledge of the written language the program is presented in is even essential.

Continue to next page