ChatGPT Prompt Engineering For Life Science Marketing

I recently watched Deep Mind’s free class on AI ChatGPT Prompt Engineering for Developers. The class provided a great primer on ChatGPT and offered helpful tips for prompts. As life science content marketers, diving too deep into the details may not be necessary, but having a more comprehensive understanding of the technology can give you greater control and help you use these tools more effectively. This is just the beginning, as there are many more tools and innovations to explore in the coming years.

At the very basic level, what’s most important is knowing what words to use in the prompt and then understanding the output that comes back.  In fact, there is a new type of job that AI is opening up, ‘Prompt Engineering’ who are tasked with the training of LLMs to continuously give users accurate and useful responses.

ChatGPT can be a really great productivity boost. The advice of ChatGPT is incomparable with the creative demands of life sciences but the very act of ‘discussing out loud’ about your ideas helps you figure out what you want to say next.

Using ChatGPT to strengthen and verbalize possibilities is probably the most important of them all.  The information and bias that large language models often produce can kick off great discussions. The fact that it’s not perfect means this is a great time to learn!

Here are my key insights from the course:

Although it is true that large language models are not always accurate, this presents a great opportunity to learn. Rather than viewing artificial intelligence (AI) as an assistant that merely provides a list of tasks, it can be seen as a thought partner. Unlike a traditional assistant, AI enables dialogue and the challenging of assumptions.

By embracing AI in this way, we can more fully explore its capabilities and discover new ways of thinking. So let’s embrace this unique opportunity to learn and grow with AI.

1. Write clear and specific instructions

• Crafting a prompt that is both clear and concise can be a balancing act. While brevity is often desired, a longer prompt can offer additional context and clarity that may lead to more nuanced and relevant responses. It is important to consider that precision may sometimes require a lengthier prompt.
• I recommend exploring different prompt enhancer tools within the ai-content generating apps and experimenting with them to discover the one that meets your needs the best. With persistence, you’ll find the tool that works best for you, providing high-quality results for your writing needs.

2. Give the model time to “think”

• To avoid an incorrect conclusion from a model, try to reframe the query instead of seeking a final answer. Requesting a chain or series of relevant reasoning increases the likelihood of an accurate response. 
• If you assign a model a task that is too complex to complete within a short timeframe or with few words, it may formulate an inaccurate guess. This is something that could also happen to a person who is asked to solve a challenging math problem without sufficient time to analyze the question. 

3. Craft engaging prompts via iterative development

• If you give a model a question that’s too technical right from the get-go it may not provide the best answer or even worse, it may make up a guess which is likely to be incorrect so it’s better to take an iterative approach 
• There’s no such thing as a perfect prompt. What matters most is the process for getting to the prompts that work for your application. 
• If you look at training for prompt engineers – it’s not about asking random questions, its more about engineering prompts to get unique responses and perspectives. Prompt engineers need to learn how to analyze prompt performance and collaborate effectively with researchers, designers, editors, and advertising managers.

I wanted to test out part 3 and see the results and see if taking the time to do an iterative approach was worth the time and effort. 

Case Study Example:

Say you want to do research to write a blog about the role of genomics and bladder cancer that lead to new diagnostic or therapeutic strategies for this disease in the future.

Option 1: Prompt ChatGPT directly

Ask “How might ongoing research in genomics and bladder cancer lead to new diagnostic or therapeutic strategies for this disease in the future?”

ChatGPT output:

Ongoing research in genomics and bladder cancer has the potential to lead to new diagnostic and therapeutic strategies for this disease. Here are some possible ways this could happen:

1. Identification of genetic markers: Researchers are studying the genetic changes associated with bladder cancer to identify specific genetic markers that may indicate the presence of the disease. These markers can be used in diagnostic tests to detect bladder cancer at an earlier stage.

2. Personalized treatment: Genomics research is also helping to identify specific genetic mutations in bladder cancer that may be targeted with new therapies. This personalized approach to treatment could improve outcomes for patients by providing more effective and targeted therapies.

3. Immunotherapy: Recent studies have shown that bladder cancer is responsive to immunotherapy, which is a type of cancer treatment that helps the immune system recognize and attack cancer cells. Genomics research may help identify which patients are most likely to benefit from immunotherapy and what specific targets the immune system should be directed towards.

4. Biomarkers for prognosis: Researchers are also looking for biomarkers in bladder cancer that can be used to predict a patient’s prognosis. This information can help doctors determine the best course of treatment and monitor patients more closely for recurrence.

Overall, ongoing genomics research in bladder cancer has the potential to improve diagnosis, treatment, and outcomes for patients with this disease.

Option 2: Take iterative approach to prompting 

Prompt #1: What is genomics?

Prompt #2: What causes bladder cancer?

Prompt #3: What role does genomics play in bladder cancer research?

Prompt #4: How might ongoing research in genomics and bladder cancer lead to new diagnostic or therapeutic strategies for this disease in the future?

ChatGPT output:

Ongoing research in genomics and bladder cancer has the potential to lead to new diagnostic and therapeutic strategies for this disease in several ways:

1. Identifying new molecular targets: By analyzing the genomic profiles of bladder cancer cells, researchers can identify specific genes, proteins, and signaling pathways that play a role in the development and progression of the disease. This information can be used to develop targeted therapies that inhibit these molecular targets, which may be more effective and have fewer side effects than traditional chemotherapy.
2. Developing personalized medicine approaches: Genomic analysis can also help identify subtypes of bladder cancer with different genetic and biological characteristics. By tailoring treatment to the specific genomic profile of an individual patient’s tumor, personalized medicine approaches have the potential to improve treatment outcomes and reduce toxicity.
3. Improving diagnostic accuracy: Genomic analysis can also help improve the accuracy of bladder cancer diagnosis by identifying specific genetic changes that are characteristic of the disease. This information can be used to develop more accurate diagnostic tests, such as blood or urine tests, that can detect bladder cancer at an early stage.
4. Predicting treatment response: Genomic analysis can also help predict how individual patients are likely to respond to different treatments, allowing for more personalized treatment selection and better treatment outcomes.

Overall, ongoing research in genomics and bladder cancer has the potential to revolutionize our understanding of this disease and lead to new, more effective diagnostic and therapeutic strategies that improve patient outcomes and quality of life.

Final thoughts:

I thought option 2, taking an iterative approach, added a more accurate, more well written and comprehensive response versus just asking the main question right away.