With growing digitisation, there’s no shortage of information you can find on the internet. But that’s led to information overload for many. Digesting, understanding and then distilling huge amounts of information for a specific purpose on any topic is now the challenge. And that’s where large language models such as ChatGPT come in.
In case you’re just catching up on this ongoing technology trend, lets step back for a moment. ChatGPT is one of several large language models including Meta AI’s LLaMA, whose name, although it curiously conjures up images of a South American pack animal, is in fact an acronym for Large Language Model Meta AI. There’s even been Alpaca, although this has since been withdrawn.
ChatGPT though, has reached new heights of popularity since its launch at the end of November 2022. It’s capable of doing some pretty clever things with your text data, and can make light work of producing text outputs based on the inputs you give it.
Large language models (LLMs) are trained on vast amounts of data, and they are designed to make it easy to adapt to a multitude of different tasks. They use machine learning and algorithms to process and understand natural language. These models use massive amounts of text data to learn patterns and relationships within language.
So why is ChatGPT causing a stir and making us sit up and give it a go?
Well from a technology point of view ChatGPT uses a transformer-based language model that is designed to generate human-like conversations. It is trained on a large corpus of conversational data and can generate coherent and natural-sounding responses to user input. ChatGPT can be used to power chatbots, virtual assistants, and other conversational applications.
So, what’s all the excitement about?
Well, it really is very easy to use and in case you’ve yet to try it, for the moment it’s free, so anyone can give it a go, which is why colleagues are dropping it into conversations and why ChatGPT is getting quite a bit of airplay. Most self-respecting students will probably be very familiar with it, with reports on the internet about how your twenty-something son or daughter may already be using it to speed up research work at university, or to produce essays, as well as using it in their first roles at work.
In actual fact, it’s less to do with the technology and more to do with our degree of readiness.
It’s not so much that the technology is new, but rather that it’s the right tool for the times. We’re all looking at tools which can help us to work smarter and faster and be more productive and ChatGPT is highly accessible and very easy to use.
This is all very interesting, but how can I use it within my organisation?
As mentioned previously, ChatGPT is trained on large public data sets including everything you can already access publicly on the internet. But what if you want to use data that you hold on your intranet or which is stored locally on staff devices? Then it’s a matter of getting ChatGPT to interface with those datasets and taking the right steps to use it effectively.
A small team at Redcentric have taken on that challenge and we are exploring the potential it offers us as an organisation. We’re developing a centre of excellence and working at understanding how we can use ChatGPT using a defined set of internal data to test out our strategy. . We’re hoping we can guide and assist our customers who may also be exploring how to use it.
Large language models can be compared to the introduction of the internet. The technology will constantly evolve and technically it requires specialist skills to refine and develop, and the constant evolution in the technology is part of the challenge. In fact, those skills are now being referred to as ‘prompt engineering’ and we’re fortunate to have a team of cloud engineers who have the skills to understand how to go about implementing these technologies.
Like any new technology there is always a learning curve. Not least of which includes that well known issue – rubbish data in, equals rubbish out. It also goes without saying that you need to ensure that you’re not using confidential IP. So far though, we’ve extracted our selected data set, put it into a content management system to curate it, clean it and tidy it up, then we’ve overlaid ChatGPT to reference that data, and we’re trying it out with a few tasks.
The results have been good so far, and we’re confident that in time we’ll be able to perform the same routine work in a quarter of the time, potentially freeing up our staff to focus on other things, and releasing time so they can respond quicker and provide better customer service to internal and external customers.
If you’re thinking about how you can use ChatGPT in your business, contact us to find out about how our team can help you to architect a solution in your business and we’ll share our learnings with you. In the meantime, here’s our top 5 tips for getting the most out of ChatGPT.
1.Curate your data – rubbish in, equals rubbish out.
Don’t expect ChatGPTv3 to interpret half-formed notes, but ChatGPTv4 does claim to handle this much better. You’d be wise to start the process of organising and cleaning up the data you plan to use.
2. Be selective when you choose what corporate data you want put into your data repository – avoid confidential IP.
ChatGPT is owned by OpenAI, who joined forces with Microsoft, who is now a key investor. It has been trained on massive public data sets, but just as you would with any third-party application, be cautious about sharing confidential data.
3. Take time to learn how to use it
Using the analogy of a google search on the internet, when searching for information you need to use the right searches and then you’re presented with the most relevant results. Learning how to instruct ChatGPT effectively is also important, different instructions worded in different ways will produce different outcomes. You’ll need to experiment to get the optimal outcome.
4. Always review
ChatGPT can produce some impressive results, but make sure you’re familiar with your data source and its limitations as well as its strengths. Always review your outputs and be aware of how your source data may influence the final output.
5. Beware of text length limits
Large Language Models like ChatGPT work using the concept of tokens, with one token resembling one syllable. ChatGPT uses a limit of 4000 tokens to process the input and create the output. So, a text of 2000 tokens will produce an output text of 2000 tokens too. In ChatGPT4, the tokens have been increased to 8000 which is a big step forward, and future versions will go further. But for now beware that if, for example, you want to request that ChatGPT writes a long guide, its best to make iterative requests for a series of different topics within the guide to get four chunks of reasonable text length to create your final guide.
For now, it’s still early stages, but even at the experimentation stage, it is proving to be great fun and very useful.
Disclaimer – this was not written by ChatGPT!