ChatGPT is a state-of-the-art language model developed by OpenAI, and it's taking the world by storm. 🌎
But, as a researcher, you might be wondering how this technology can help you. 🤔
Time is precious in the research world, with a never-ending list of tasks to complete, from writing research papers to conducting experiments and analysing data. And that's where ChatGPT comes in.
It’s your Personal Research Sidekick! 🦸🏻
With lots of contention around this radical new chatbot, let’s dive into the benefits and risks of using ChatGPT in science.
We will discuss:
How ChatGPT can improve and speed up your scientific writing tasks.
The drawbacks and risks to be aware of when using ChatGPT.
Tips for getting the most out of ChatGPT.
ChatGPT: Revolutionising Research
To demonstrate the capacity of ChatGPT, the first thing we did was to ask ChatGPT to write the premise of this blog post. 😉
And well, you tell me… Did you notice that this was actually written by artificial intelligence (AI) 🤖 rather than by a human?
It’s so elegantly written!
Now think of all the ways you could apply this to your research and writing tasks! Think of all the time you would save! 🙌🏼
So… what is ChatGPT?
If you haven’t already heard of ChatGPT (Generative Pretrained Transformer), then listen up!
This mind-blowing chatbot from Open AI was released on November 30th 2022 and has already amassed an impressive 100 million active users within the two months since its release. 🤯
So what makes this AI so special?
Well to start with, ChatGPT is currently FREE and easily accessible for all users (for now).
And from the perspective of AI, this model is a game-changer. With an impressive capacity to understand and generate coherent, text that mimics human language, it’s no wonder this AI is taking over. 📝
And my favourite part - it’s SO EASY and intuitive to use! You can chat with it just like you’d chat with a human being. It can respond to follow-up questions, admit mistakes, challenge incorrect information, reject inappropriate requests, and, like a human, remember previous requests. 🤯
The advanced language capabilities of ChatGPT are made possible by a neural network called the Transformer architecture. Neural networks, which are inspired by the human brain, allow ChatGPT to learn and mimic language patterns and structures. To accomplish this, the network was trained on a massive dataset of text from the internet, with human trainers moderating outputs.
In other words, it’s just a really smart AI that is set to revolutionise just about every field. 😉
How can ChatGPT help researchers?
Let’s jump into the good stuff: how can this revolutionary AI model help you?
In a nutshell, ChatGPT can help you overcome writer's block and turbocharge your writing process. 👏🏼
The options are endless:
Formulate emails and replies 📧
Write a stellar researcher bio 👩🏻🔬
Simplify complex topics 🧪
Brainstorm ideas for a grant application ✏️
Compose social media posts 💬
Suggest analogies for your research (hello 3MT!) 🚙
Fix your grammar 📃
Translate into a variety of languages 🇬🇧🇧🇷🇫🇷🇪🇸
And SO MUCH MORE!
Every time you open it up, you’ll find a new use for it.
It is like having a really intelligent friend that knows all of humanity's knowledge and has nothing better to do than to help YOU! 😉
Perhaps you have to write an apology email? Ask ChatGPT to do it for you! 📧
Need some help starting off your literature review? Ask ChatGPT for some openers for you! 💬
So what do you need help with? Why not ask ChatGPT!
But beware: use ChatGPT with caution!
Whilst ChatGPT is a powerful tool, just like all the other SciComm tools in our toolbelt, 🔨 that can boost the writing process for researchers, it is not without its own caveats. Let’s break down some of the pitfalls causing widespread concern.
Risk 1: Inaccuracy
This is probably the biggest downfall, particularly for scientists, and one you should not forget!
ChatGPT can easily produce plausible-sounding but completely incorrect, misleading responses.
Like this one. 😂
And yes I checked… Dr Tullio Rossi award-winning physician in Milan does not exist! 😂
This kind of nonsensical reply is often generated when it comes to technical topics with limited training data available. Also, the current model (ChatGPT-3) is trained on data up to October 2021. This means that it has no clue about recent events.
In science, facts and accuracy are crucially important.
So be mindful, always fact check and do not trust ChatGPT for accuracy!
Risk 2: Unreliable References
The model's unreliability means that it cannot be trusted to provide accurate or authentic references and citations.
They may look real, but in fact none of these scientific articles exist! So we would suggest that you avoid asking ChatGPT for literature recommendations.
Risk 3: Safety and legal concerns
While ChatGPT is an incredibly powerful tool, there are several potential safety, copyright and ownership concerns to consider.
One concern is the potential for the technology to inadvertently generate offensive, racist or inappropriate content. 😮
Another major concern is the legal implications of using ChatGPT and other AI-generated content.
Who owns this content?
It has been ruled that artwork generated by AI, such as DALLE, cannot be copyrighted.
Will the same be true for text?
A number of scientific journals, including Science, have already banned the use of AI-generated text within scientific articles. 🛑 Springer-Nature, have also expressed concern, laying down two rules to govern the use of AI in assisting with manuscript preparation, stating that:
ChatGPT, and others, will NOT be accepted as a credited author on a research paper.
Researchers must disclose the use of such tools in the methods or acknowledgements sections.
Essentially, these tools should only be used to improve the readability and language of a research article, not replace the interpretation of data.
It is important to understand that this technology is still so new that the rules surrounding its use and legal implications are still being written.
Risk 4: Training Bias
As ChatGPT has been trained in a certain manner, the model has been exposed to the biases and inaccuracies present within the training data. Imagine a kid raised by wolves; they know nothing outside how they were raised.
As the human trainers who moderated ChatGPT’s output had a preference for longer, more comprehensive answers, the model responses are excessively verbose as a result.
Tip: Ask ChatGPT to rewrite the response concisely. 😉
The phrasing of the input can greatly impact ChatGPT's response, so providing some clear input is crucial to avoid ambiguous interpretations. Slight rephrasing of a question can lead to accurate answers, while ChatGPT may guess the user's intended meaning if the input is unclear.
Unfortunately, ChatGPT has yet to learn to read human minds.
Also, given the way that ChaGPT was trained, you can trick it into thinking it was wrong. Just like this Tweet that went viral. 😂
The good thing is, as a scientist, once we know the limitations of a tool, we can work around them.
To save you some time we have optimised the use of ChaGPT for you, just like you would for new lab equipment 🔬! Read on for some of our tips on getting the best out of it. 👇
How can I get the most out of ChatGPT?
Here are our top 3 tips for formulating a good input prompt: ✨
Tip 1: Use descriptive prompts.
When using ChatGPT, vague or ambiguous prompts may lead to irrelevant or unhelpful responses from the model. Instead, ask focused questions that require straightforward answers.
Also describe what type of response you would like. For instance, ask ChatGPT to summarise this paper in a way that a 12-year-old would understand. Defining the level of formality, the tone, and the language ChatGPT uses can help ensure that the responses match your desired communication style.
Here are some simple examples; use as many as are relevant to create a descriptive prompt:
Tip 2: Provide context
Providing context is crucial when communicating. Full stop!
ChatGPT is no different.
Take advantage of ChatGPT’s ability to understand context and maximise the chances of a relevant response. A key strength of ChatGPT is that it can remember previous statements or questions in the conversation.
So try building on previous statements and ask follow-up questions to get the most out of ChatGPT. Be sure to provide all relevant details and background information.
You can do this in a number of ways:
You can even tell ChatGPT "you are now a confident marketing expert with 20 years experience, explain to me how to boost our social media engagement"... What?! 🤯
You may have noticed at the beginning of this post, that the input we gave ChatGPT to write the premise of this blog didn’t look all that detailed…
Well, that’s because we built up to it. Here’s how we did it:
We first provided a link to our previous blog 5 acting lessons to level-up your public speaking skills and asked ChatGPT to mimic the friendly, active voice used.
Next we described the content of the blog using keywords in quotation marks, and asked ChatGPT to write a premise.
The resulting output was okay, but we knew we could do better! So we asked it to re-write using the ‘And, but, therefore’ template of storytelling.
Context is essential for accurate and useful responses!
Tip 3: Use clear and concise language
Using simple and direct language is key to getting the most useful responses from ChatGPT.
When formulating your prompts, avoid using too much jargon, or convoluted sentences that could confuse the model. Instead, focus on communicating your question or prompt in the most straightforward way possible, while still providing the necessary information and context required.
Along the same line of thought, typos and grammatical errors can cause ChatGPT to misinterpret your question, leading to irrelevant or inaccurate responses. So be sure to proofread your input so that ChatGPT clearly understands your message.
Fun fact: You can always edit your prompt! So if you find an error, or don’t get the desired response, tweak your prompt and try again. Unlike your lab mates, ChatGPT has an unlimited supply of patience!
Extra Tip: Always fact-check
Okay, so we know we kind of mentioned this in the risks section, but it is just that important!
Fact check. Fact check. Fact check.
In order to verify the accuracy of the information provided we beg and plead for you to always do your own research. Use Google! Read the literature! Check your old notes! Ensure for your own credibility that the information is correct before you use it.
As long as there is human oversight to fact-check the details, ChatGPT can be a very useful tool.
Where to next?
There is no denying that ChatGPT is a revolutionary move forward in the world of AI. It can speed up writing tasks such as papers or grants, provide inspiration and direction or answer a quick question.
But the question on everyone's mind… well what next? 🤔
As generative AI continues to evolve and advance, the possibilities for its application in science are becoming more apparent.
The computer science behind generative AI is moving lightning fast, ⚡️ with the release of OpenAI's GPT-4, expected later this year. It is anticipated to be more sophisticated, built on a larger data set, with specific tools tailored for particular industries, such as academia or medicine.
The recent development of Med-PaLM, a clinically-focused generative AI by Google and DeepMind, demonstrates the potential for AI to be given field-specific training. This model, although it still had shortcomings and unreliability, was able to answer open-ended medical queries almost as well as human physicians.
Similarly, ChatGPT passed the gold standard medical exam in the UK, and is capable of diagnosing certain conditions. This means that, in theory, its knowledge base rivals that of a trained medical doctor. 🩺
New AI innovations appear every month, making it hard to understand how these tools will affect the future of science and technology.
AI is like the Wild West right now.
We can only hope that regulations keep pace with new developments and ensure that AI is used in a safe and ethical manner.
‘Excited’ and ‘apprehensive’ are the words being used to describe ChatGPT in Science.
Whilst the avenues for generative AI models, such as ChatGPT, are limitless, there is also widespread concern on how to best use these models (if at all). We believe there can be a place for these such models to enhance and accelerate writing in science, if used with good judgement.
But we need to remember:
ChatGPT is a trained model, with its own biases and limitations.
You are the expert and ChatGPT is just a helpful tool.
So where does that leave us?
Machines can only go so far. And accuracy matters in science. That’s where we come in!
Our team of PhD-trained science communicators and creative experts at Animate Your Science leverage their scientific expertise to curate captivating stories and turn them into visually stunning and easy-to-understand scientific animations and graphics. By showcasing your research in animated form, you can be sure to capture the attention of your target audience and stand out in your field.
Let us help you bring your research to life and make an !
Click here to learn more about our services.