You know how people call all vacuum cleaners “Hoovers”, even if they’re not made by Hoover? Well, that’s kind of what’s happening with ChatGPT and large language models or LLMs. Just like Hoover, ChatGPT has become so famous that people think of it anytime they hear about LLMs. But there are many other LLMs out there, all with their own special skills in creating realistic text and having smart chats. In this article, we’ll look at ChatGPT’s rise to fame, introduce other LLMs, and encourage you to discover more about AI.
ChatGPT using GPT 3.5 was released by OpenAI in November 2022, and people immediately noticed how good it was at having conversations. OpenAI, a company focused on making AI helpful for everyone, built ChatGPT using a method called generative pre-trained transformer (GPT). This approach helps ChatGPT produce very realistic responses, making it useful for all sorts of things, from helping with code to writing essays, and even for having therapeutic chats.
LLMs like ChatGPT create text that sounds very human, but they don’t really “get” language or the world like we do. They work by finding patterns in data and copying those patterns. During their training, they are given lots of text data. They learn the statistical patterns in this data, like which words tend to follow others. When you give a model a prompt, it creates a response by figuring out which words or phrases are statistically likely to follow your prompt.
Take the sentence “It’s raining, so I’m taking my…”. An LLM is likely to finish it with “umbrella”. But it doesn’t understand that rain is wet or that an umbrella keeps you dry. It just knows from its training data that “umbrella” often comes after “rain”. It’s very good at predicting patterns, but it doesn’t really understand them. This shows why we need to be careful with LLMs. They can be misunderstood or misused, so it’s something to bear in mind when you’re testing or choosing an LLM.
At the moment, GPT-4 is the leading LLM. But you can only use it through an API or if you pay $20 a month for the ‘Plus’ service. Bing Chat, which is based on GPT-4, is quite different. Instead of creating text, it mainly searches the internet and simplifies the information it finds.
But there’s more to LLMs than just ChatGPT. There are many other models, each with their own skills and weaknesses. You can try these out on websites like Poe.com. For example, the LLM ‘Claude’ is known for its vast knowledge base, while ‘Sage’ is good at dealing with languages other than English. Trying different models is the best way to get to know what they can do.
A new LLM called ‘Claude 2’ has been developed by Anthropic. While it doesn’t reason as well as GPT-4, it has one big advantage: it can consider more information at once. This makes Claude 2 really good at analyzing and responding to large documents.
Then there is Google’s Bard which is way behind even the original ChatGPT but is catching up fast. With its huge resources, it will no doubt be a contender in the future but for the moment, ChatGPT with GPT is model to beat.
ChatGPT’s success has made LLMs popular and exciting. But there’s a lot more to discover in the fast-moving world of big talking AI models. To really understand these tools and join the conversation about them, you need to try out different LLMs and find the one that suits you best. So, why not start exploring and getting involved with this amazing tech?
Photo by Sanket Mishra from Pexels