OpenAI, Chat GPT, and generative language models
2023 will go down as the breakthrough year for ChatGPT due to the intuitive capabilities and wide scope of applications of the tool.
But where did it all start?
Created in 2015, OpenAI has notable founders such as Elon Musk and Sam Altman with the aim of integrating artificial intelligence into society for the benefit of humanity. Looking back on the last seven months, you can definitely say their work has impacted society through their creation of ChatGPT.
In a nutshell, ChatGPT is an AI application that has been ‘trained’ to understand natural language and conversation. Pulling data from the internet and presenting its findings in easy to understand responses. Furthermore, it can be used as a translator, create content, summarise text and process / write code.
The R&D
Ironically, you can find out about the R&D behind ChatGPT by simply asking it. The tool identifies five components that make up its development:
- Training Data – Large amounts of text, sources that include books, articles, and websites – where it can capture grammar and context.
- Pre-Training – Learning to predict the next word in sentences and understand the patterns and relationships it sees from its training data.
- Fine-tuning – Final stages of learning translation, question answering, and conversation.
- Architecture and Algorithms – Transformer models help natural language processing tasks, for example, capture contextual relationships effectively.
- Iterative Development – The final stage of testing the previous four components, making fine adjustments to any imperfections.
However, it can be difficult to put all this development into perspective and understand the gravity of its intelligence. The English language is comprised of one million words. In comparison, ChatGPT is made up of ten billion words, including fifteen languages (incl. English, French, Russian, Chinese, Arabic and Urdu), and sixteen programming languages (incl. Python, Java, JavaScript, C++ and, HTML/CSS).
If you wanted to dive deeper into the development of the tool, tokenization is the breaking down of the sequence of text into smaller units, appropriately called tokens. These help ChatGPT’s natural language processing (NLP) for further development.
Applications
As previously mentioned, OpenAI reached their goal of impacting society through their work. The applications for ChatGPT are vast, content creation, consultancy, and customer support to name a few.
A popular term being used more and more within industries is, ‘ChatGPT won’t take people’s jobs, people that use it will’. Which to an extent is correct. If utilised correctly by competent professionals with in-depth knowledge of their profession, then ChatGPT can allow for streamlined workflows and increased productivity.
Despite having ten billion words, mistakes can still be made. Users who copy and paste its responses risk not only incorporating the errors the tool might have created but also limiting themselves by hindering personal development in their respective careers.
Introducing MarIA
FI Group has made the decision to move away from ChatGPT. At a glance this decision might undermine what was said previously about streamlining workflows, however, we have opted to use a different tool. We are launching MarIA which uses AzureGPT functions. Similar to ChatGPT, but can ensure that conversations and data shared with it remain in the Azure environment which provides an additional level of security.
On the topic of ChatGPT, FI Group France Scientific Director Charlie Grosman said:
‘while it generates useful and relevant information, in a short period of time, it is not a technology that employees should trust at an exact level».
What does MarIA mean for our clients?
FI Group requires a lot of sensitive information from our clients such as confidential R&D project details and employee salary, if this information is shared with ChatGPT it is then stored in OpenAI’s database and outside of our control. MarIA ensures that any private information is kept safe in our own database and cannot be accessed by anyone outside of FI Group.
Moreover, MarIA will also limit how much FI Group employees can use its functions. Ensuring that work is being done by real people and aided by the tool, when necessary, rather than the other way round. Meaning, clients can be assured they are getting genuine expert client opinions.