ChatGPT Essential Guide for IPR (with practical examples)
ChatGPT is like the world's most knowledgeable and smartest intern available for free (GPT 3.5 model).
Then why are many people unable to achieve good output with ChatGPT?
There are two reasons for that:
Reason 1: Providing a generalized prompt
When we give it a generalized prompt, it gives us a generalized response. It's not because it is incapable, but because it "thinks" this is what you want. We will learn how to write a detailed, task-specific prompt today.
Reason 2: Not making use of contextual memory
Many people use ChatGPT only for one-shot prompting i.e., providing only one prompt and taking only one response. ChatGPT maintains a contextual memory of the previous conversation. We will learn how to leverage that today.
Let's address the above two points in detail:
- Utilizing Prompt Engineering for specific prompts: Prompt Engineering is a subtractive process. It is not an additive process. What does this mean? This means that by making a detailed prompt, we are not increasing ChatGPT's scope of relevant knowledge because its scope is already the entire world. Instead, we narrow its focus to obtain more tailored and suitable responses by providing some constraints. Below are four examples illustrating this, all using the free version of ChatGPT.
- Leveraging Contextual Memory: The 4 examples shown will be a part of a single conversation where we will make a call back to the previous conversation to get a better response.
Example 1:
Constraints in this example:
a. Research assistant – By writing this, we gave it a persona and tuned it to think like a researcher.
b. 150 words – It has to give an answer in only 150 words.
c. Focus on – Constraining what it should focus on while answering.
Example 2:
Constraints in this example:
a. Explain in simple language – We are limiting it to keep the explanation simple. Alternatively, we could write “Explain in detail” depending on our use case and understanding level of the subject matter.
b. Go step by step – By adding this constraint, we are telling it to explain in a step-by-step way, rather than explaining in a single paragraph since the former is more readable.
ChatGPT Response:
Example 3:
It is difficult to understand a patent's innovation just by understanding its broadest independent claim, right? Which is why we personally love this prompt.
Constraints in this example:
a. Relevant help – In this example, we have given it multiple embodiments of the patent (specifically, text for figures 7 to 12 since they were the most relevant ones) so it can explain the patent claim again in a more detailed and specific way.
ChatGPT Response:
It is also worth noting that we are making use of ChatGPT's contextual memory here by making a call back to the previous conversation – by writing "Explain the previous patent claim again".
Example 4:
Constraints in this example:
a. 10 companies (with product names) – We are asking for 10 products from 10 different companies which we can further explore for infringement analysis.
b. US-based – By writing this, we are limiting the scope to US companies only, this also reduces hallucinations and increases factual accuracy.
c. Similar technique – We are asking it to look for companies that provide a similar technique as the previous patent claim.
d. Not exact matches – Sometimes GPT says that it can't provide real-time information and so on, we are telling it that it's fine to tell us similar companies only, they don't need to be necessarily infringing on our patent claim.
Once again, we leverage contextual memory by referring back to the previous conversation.
In conclusion, by refining prompts and leveraging contextual memory, we unlock ChatGPT’s full potential for IPR and beyond. Embrace these techniques for insightful outcomes and boundless possibilities. Happy prompting!
