Our opinion on new tech in Air Cargo

What will be the main driver in the Air Cargo industry in the upcoming years? To us, BeCon Projects, as Air Cargo consultants, the answer is clear: new tech is the driver in Air Cargo.  However, what does that mean? Well… everybody talks about OpenAI these days and as consultants, we see it as our duty to keep up to date with new technological trends to add value to the industry (check our blog post covering TIACA’s article to understand our business model under How BeCon Projects As Air Cargo Consultancy Adds Value). 

So, let’s get the facts straight first: ChatGPT was the first chatbot launched by OpenAI in November 2022 It uses a GPT model in which you could easily ask questions via a prompt to get answers from it based on its knowledge and data access. (ChatGPT – Wikipedia)

 

For those of you who have not heard of a prompt: An artificial intelligence (AI) prompt serves as a way to interact with the language model of the bot easily, it does not have to necessarily include a question, a human could also interact with a bot via code extractions or simple text.  

 

What ChatGPT can do

ChatGPT would use its knowledge to help answer your question in a humanly understandable form. It can generate code or write poems, songs, emails, essays, etc. (see more here: Introducing ChatGPT (openai.com)). It got criticized a lot because school kids used it to do their homework for them. I am sure some parents, who read this blog, have had the occasional debate about this at home…

So far so good, but what versions of it are out there and how does one access them? To use its most recent models, one has to do it via the chat completions API endpoints. The first version, GPT-3.5, which is free and known to most people, is limited to 2021 data. By the way, as I am very into tech topics, Chat GPT-3.5 has a data set of ‘300 bio words and 570 GB’ according to Aparna Iver in her article Behind ChatGPT’s Wisdom: 300 Bn Words, 570 GB Data (analyticsindiamag.com), published from Dec 2022! Can you believe that? These 300 bio words used come from sources such as ‘books, news, and articles’ as explained by Jeff Joyce in his blog which posted Where Does ChatGPT Get Its Data? – Content @ Scale (contentatscale.ai), released in February 2024.

Therefore, it provides a huge access to knowledge, making it a powerful tool! The well-known version of ChatGPT used publicly available text, accessible on the internet. Humans only contributed to its success and accuracy by interacting (done by AI trainers) with users and AI assistants during conversations as you can (GPT-3.5 Turbo fine-tuning and API updates (openai.com)). 

 

Limitations of chatbots such as ChatGPT

However, chatbots such as Chat GPT have some limitations. They are prone to hallucinations, a term used to describe that bots can make up information, as illustrated by Cade Metz, who wrote in the New York Times about them. Cade Metz, however, also explains that the lowest rate of hallucinations according to research done by the start-up Vectara would be found at Open AI. There, it would happen ‘around 3 percent’ of the time (https://www.nytimes.com/2023/11/06/technology/chatbots-hallucination-rates.html).

 

 

I already talked to an Air Cargo industry expert at the HOLM conference last year, where he told me that he believes simple tasks can be automated via bots such as ChatGPT. Examples of this would be Excel templates, which have to be created in the right format, or routine tasks that suit back-office activities. Coming from a software company previously, I remember former colleagues telling me that code coming from chatbots must be rigorously checked before implementing it. Hence, I believe that chatbots such as ChatGPT currently cannot replace creative and highly complex tasks such as writing difficult code that must be customized and embedded securely into a developer’s repository.

 

It can, however, serve as a good orientation of how code could be written and give us food for thought on how to improve our everyday work, as a developer, in the back-office, or in jobs related to producing content such as Marketing.

 

The functionality and power of AI

In a nutshell, ChatGPT has shown just how powerful AI can be. I just love how it uses NLP (Natural Language Processing) to help digest the input (from a prompt).

Firstly, ChatGPT will use NLP tech to process the message you have put in. It then cuts it into smaller “bites” (units), to subsequently turn them into a numeric representation. Secondly, the numeric representations of your initial message, which has been cut into smaller “pieces”, are then passed on to the model whose job it is to learn to comprehend your query. To make it very practical, ChatGPT is a neural network, which, as stated above, was trained beforehand based on a lot of textual data. This training, which has happened previously, enables the model to literally predict the next word in a sentence based on the words used beforehand. This is precisely what enabled ChatGPT’s model to use grammar and present knowledge in a humanly understandable way. To do so, it then takes its numerical representation and “reverts it” back into human text form (Natural Language Processing in ChatGPT: An In-depth Exploration – Jenni AI).

 

Since it is such a powerful and intelligent tool that enables answers to complex questions and allows us to have conversations, write automation, and correct code, we will, for sure, see more versions of “company’s own chatbots”. One good example is the one used by Lufthansa Technik, who has also presented their AI solutions at several events we attended such as the HOLM conference last year (Start now with AI in your company easily (lufthansa-industry-solutions.com)).