The ChatGPT chatbot has caught on in a big way. So, what makes this AI chatbot different and why is it so important?
What exactly is artificial intelligence chatbot ChatGPT?
It’s safe to say that ChatGPT has exploded since the start of 2023, but it was actually released at the end of November 2022 by OpenAI. For those who don’t know, OpenAI was originally a non profit AI company co-founded in San Francisco by Elon Musk, but is now under the control of Microsoft.
ChatGPT stands for ‘Chat Generative Pre-trained Transformer’ and while the prototype launched last year, it’s really in the first month or so of 2023 that it has caught on across the world. Since its launch, OpenAI has been valued at around US$29 billion.
How do large language models like ChatGPT work?
ChatGPT has gone through several iterations of AI development and is built on top of GPT 3 through the use of reinforcement learning and supervised learning.
Technology like this is obviously incredibly sophisticated and relies on human trainers to improve and refine its performance. Supervised learning involved the human trainers providing both sides of the dialogue, which is then ranked and given a reward model.
These help the system to ensure it’s as fine tuned as possible, and it used more iterations including Proximal Policy Optimisation (PPO) to do so. Now that ChatGPT is in regular use by millions around the world, it continues to collect data that can be used to further refine its language model and the AI system in general.
The users can grade the response they get from ChatGPT and answer questions to give feedback on the quality of the answers they got from the Microsoft-owned AI program. The search engine Bing has integrated ChatGPT, so that it’s simple to use.
What can ChatGPT do?
While on the surface ChatGPT is a conversational chatbot, its potential is far more wide reaching. Some people are already speculating as to whether ChatGPT or other similar AI tools may even entirely replace the Google search engine.
Google itself has responded to ChatGPT by announcing its own competitor called Bard, which is expected to be released in the coming weeks (as at February 2023).
Even for users who are very much not into following tech trends or other applications, ChatGPT has proven tempting to at least try. You can simply type in a question and ChatGPT responds and writes plausible sounding answers.
In addition to supplying (sometimes) correct answers, the artificial intelligence tool can also write student essays, summarise documents and even write software. It can be used as an alternative to Google as it sweeps the Internet for answers to questions, and as it’s generative AI, the expectation is that it continues learning.
Is ChatGPT free?
Yes, for now, anyone can simply type their question and have a conversation with this version. By giving it a command and prompts, you can receive all kinds of answers, essays and streams of language-based consciousness. You can also use the technology to write code and access information from across the Internet.
ChatGPT will remember the main points of your conversation and use this to create its next answer/response. New features are added all the time as it continues to refine its responses using Google and many of the most common applications already in use.
But how accurate is it? Can tools like this really replace the knowledge displayed across the Internet and words written by human beings? The answer to this is no, at least not yet.
How accurate is ChatGPT?
Given everyone can access the tool, there are concerns surrounding inappropriate requests. Ethical concerns abound from this, given that ChatGPT will use training data and its limited knowledge to formulate its future answers.
It is incredibly creative and there’s no doubt that its commentary and answers surrounding world events, for example, can sound very authoritative and convincing. OpenAI itself has been honest about its many limitations, however, and states that more context is needed by people using it.
OpenAI CEO Sam Altman said on Twitter that: “… It’s a mistake to be relying on it [ChatGPT] for anything important right now. We have lots of work to do on robustness and truthfulness.”
What next for ChatGPT?
Following its success at the end of last year, Microsoft decided to invest billions into OpenAI (this is what led to the modified version on its search engine Bing). The search engine uses the technology developed by OpenAI to process all the searches, gather all kinds of sources for results, summarize the information held in documents, answer questions and chat with humans.
This is a huge step for search engines and artificial intelligence and could feasibly change the whole way that we use the Internet in the future. However, nonsensical answers, false positives and its potential for illegal activities all have to be considered in this journey towards total AI dominance.
A huge step forward for a large language AI programme
There is no doubt that ChatGPT is hugely fun to use and you can ask it all kinds of questions that end up with very plausible sounding answers. You can ask it to write poems and essays and then ask it to make them more exciting or more interesting.
Or you could prompt it to write software that will show you how to rearrange the letters of a word, for example. Prompts and the writing that ensue certainly make it feel like ChatGPT is pretty much sentient and has an an enormous ability.
However, it’s essential to understand that the program doesn’t actually ‘know’ anything. It’s retrieving data from a vast swathe of web pages that includes everything from Wikipedia to random blogs, for example. This means it can learn wrong information and then extrapolate that into its writing.
OpenAI warns that whatever response you get from ChatGPT, it may simply be incorrect. You can use a text box to give your opinion on the response you get, which will help it continue to learn. But it is learning, and that’s the crucial element.
By February 2023, estimates say that ChatGPT already has more than 100 million users per month. While the original ChatGPT is still free, OpenAI have added a paid version that works faster and doesn’t stop working during peak times.
Just remember that, for now, the response you get from your prompts may sound like something an associate professor would say, but it still could be completely wrong.