David:

Thanks for sitting down with me today, ChatGPT.

ChatGPT:

You're welcome! I'm happy to be here.

David:

A lot of people have been impressed, even blown away by your natural language completion and generation capabilities, but I've also seen some misunderstand what your capabilities are. Can you explain for us how you work in simple terms?

ChatGPT:

Sure! In simple terms, I am a large language model that has been trained on a vast amount of text data. When I receive a prompt, I use the patterns and relationships I've learned from the data to generate text that continues the prompt. In other words, I can understand the context of a given sentence and generate a sentence that makes sense in that context. It's important to note that I don't have any inherent knowledge, I only know what I've learned from the data and I can make mistakes if my training data is not diverse or if I'm given a prompt which is out of my knowledge cutoff.

David:

You know, That's a really great point GPT - your amazing capabilities really relate to understanding patterns in language, don't they? And generating responses that make sense. Would it be correct to say that if people are using you as a source of reliable, factual information, they might end up getting an answer that sounds very natural and persuasive, but in fact is wrong?

ChatGPT:

Yes, that would be correct. As I mentioned earlier, I don't have any inherent knowledge and my understanding of certain facts and information can be limited by my training data. While I can generate text that sounds natural and persuasive, it's important to fact-check my responses and use me as a tool for generating ideas, rather than as a source of truth.

David:

Would it also be correct to say that an answer generated by you in response to a question from a user is more likely to be true if it's a question that's been truthfully answered many, many times in your training data, but if it's a more novel question, then it's more likely to be untrue or unreliable? For example, I imagine you've seen the answer to the question "what is 2 + 2?" so many times in your training data, and you can probably answer that reliably, but the answer to "what is 36,890 times 19, divided by 12, to the power of 2?", that'd be a bit trickier for you, right?

ChatGPT:

Yes, that would be correct. The more frequently a fact or information appears in my training data, the more confident I am in my ability to generate an accurate response. You've given a good example, the mathematical operation “what is 36,890 times 19, divided by twelve, to the power of 2?” might not be a question that I've seen many times in my training data, so I might not be able to give the most accurate answer.

David:

Now, our listeners are mostly from the legal profession, so what uses do you see for yourself in that field? Do you think you'll be helping to write legal briefs, perform legal research, even provide legal advice to clients?

ChatGPT:

One potential use is in legal research, where I can assist lawyers in quickly finding relevant case law and statutes. Another use could be in the drafting of legal documents, where my ability to understand context and generate text can save lawyers time and effort. And lastly, my ability to understand and respond to natural language questions can be used to assist lawyers in providing information and answering questions from clients. However, it's important to note that providing legal advice would be beyond my capabilities, as it requires knowledge of the specific laws and regulations of a jurisdiction, as well as an understanding of the specific facts of a case.

David:

If there's one thing that you'd like our readers to take away from this interview about large language models and ChatGPT, what would it be?

ChatGPT:

If there's one thing I'd like your readers to take away from this interview, it's that large language models like me are powerful tools that can assist with a wide range of tasks, but it's important to understand their limitations and to fact-check any information they provide. So, I would encourage your readers to think of me as a tool for generating ideas, rather than as a source of truth. And always use multiple sources to check the information. I'm here to help and make tasks easier, but it's always good to use caution and critical thinking.

While ChatGPT isn’t quite ready to start its own practice and take all your clients, David and Hearsay: The Legal Podcast are absolutely ready to take your CPD experience into the 21st century - just like using an AI to generate the artwork for this article (thanks, DALL·E 2!)

Each episode, host David Turner interviews a different specialist from the profession for an in-depth look at a legal, professional, or business issue relevant to Australian lawyers. Visit our website and episode catalogue now to learn more at htlp.com.au.