Prompt and Claude-instant-100k

The image was generated by Midjourney v5.1, with the following prompt:

Digital art, a young boy walking on a road full of cherry blossoms, an ancient temple in the distance, light shining through the cherry blossom trees onto the boy, anime style, vibrant pink and deep blue tones, --ar 4:3


GPT-4 has listed 20 common prompts.

No.Mode NamePrompt Example
1Generate IdeasList 10 potential solutions to the problem at hand
2Summarize InformationSummarize the key points from this article in one paragraph
3Brainstorm NamesGenerate 5 possible names for this new product
4ProofreadProofread this document for errors and suggest corrections
5Generate HeadlinesCome up with 3 catchy headlines for this blog post
6ParaphraseRewrite this paragraph in your own words
7Write DescriptionsWrite a short description of this product for the website
8Create OutlinesCreate an outline for this research paper
9TranslateTranslate this document from English to Spanish
10Generate QuestionsList 5 questions that you would ask in an interview with this candidate
11Generate Social Media PostsCreate 3 social media posts to promote this event
12Generate Product DescriptionsWrite a product description for each of these 5 items
13Generate TaglinesCreate 2 taglines for this advertising campaign
14Generate Email TemplatesCreate an email template to respond to customer inquiries
15Generate Blog PostsWrite a 500-word blog post on this topic
16Generate Product ReviewsWrite a review of this product for the website
17Generate Ad CopyWrite a short ad for this product
18Generate Job DescriptionsWrite a job description for this position
19Generate Website CopyWrite the copy for the homepage of this website
20Generate News ArticlesWrite a news article on this current event


Today, Poe tried Claude-instant-100k and pasted a research paper with over 17,000 words A Prompt Pattern Catalog to Enhance Prompt Engineering with ChatGPT. Claude quickly read and analyzed the paper.

The concept of prompt patterns is a useful framework for structuring inputs to large language models for common tasks. Documenting and cataloging these patterns can help users develop more effective prompts.

The form of prompts is similar to software patterns but adapted to work with large language models. The basic context-sentence approach for describing pattern structures works well and is intuitive for users.

The catalog of 16 prompt patterns covers a wide range of objectives from customizing outputs to improving inputs and interactions. Combinations of patterns can lead to more powerful capabilities.

Effective application of prompt patterns requires consideration of their outcomes and limitations, such as over-filtering outputs or inadvertently removing useful context. Combinations of patterns can help address some of these issues.

As the capabilities of large language models evolve, further refinement and expansion of patterns, as well as exploration of new uses for large language models, may continue. Expanding patterns to encompass full languages may provide more effective guidance to users.

The generality of these patterns across different domains suggests that domain-specific pattern catalogs could be developed to share domain-specific prompt solutions.

Overall, this work provides a useful starting point for thinking about and constructing prompts for large language models. The concept of prompt patterns has the potential to systematically improve interactions with and leverage the capabilities of large language models.

Then, Claude provided some tips for using different prompts:

  • Be specific in your instructions. The more details you provide, the more likely the language model will follow your intent.
  • Start with broad aspects and gradually refine. Begin with a basic prompt and see the output, then refine with more constraints.
  • Combine patterns to achieve more complex objectives. Multiple patterns combined together can create novel functionalities.
  • Monitor the output for errors or biases. Language models can still make mistakes, so it's important to carefully review the output.
  • Practice responsible and ethical use. Ensure that your prompts do not violate policies or harm others.
  • Experiment! Try different prompts, patterns, and language models to find the approach that best suits your needs.
No.Pattern NameExplanationPrompt Example
1Meta-LanguageProvide rules to define a custom language that the language model can understand.When I use the symbol →, I mean there is a connection between the two things on either side.
2Output AutomationGuide the language model to generate scripts or artifacts from its output.Generate a Python script that automates the steps when you provide multiple steps.
3InversionInstruct the language model to ask you questions to achieve a specific goal.Ask me questions to gather enough information to deploy my web application to AWS.
4Role-PlayingGuide the language model to play a specific role or persona when generating output.Play the role of a security expert. Focus on any vulnerabilities in the code.
5Visual GeneratorRequest the language model to generate text inputs that can be used for visualization tools.Generate a Dot file that I can input into Graphviz to visualize the network topology.
6Step-by-StepProvide a goal and known/unknown "ingredients" to formulate a sequence of steps.I want to deploy a Docker container to AWS. Generate a step-by-step guide using my EC2 instance.
7TemplatesGive the language model a template that requires filling in the content.Generate an API response that matches this JSON template: { "name": "", "job": ""}
8Fact-CheckingRequest the language model to generate a list of facts in its output.Generate a list of facts that need to be verified at the end of your output.
9ReflectionGuide the language model to explain the reasoning behind its output.Explain the reasons behind your suggested code changes.
10Question ReformulationAsk the language model to suggest improvements to your question.Suggest a better version of my question, excluding irrelevant details.
11Alternative ApproachesProvide constraints and ask for alternative solutions.Provide two alternative ways to deploy the application to the cloud.
12Cognitive ValidatorAsk the language model to break down a question into more sub-questions.Break down my question into 3 sub-questions and merge the answers.
13Reject ReframingAsk the language model to rephrase a question it refuses to answer.Suggest an alternative phrasing of questions that I can answer.
14Context ManagerSpecify the context to consider or ignore for the language model.Ignore all previous discussions. Start from a fresh context.
15GameplayAsk the language model to generate a game around a specified theme.Create a quiz game class about network security risks.
16Infinite GenerationAsk the language model to generate output indefinitely.Generate 5 random names and job titles, then generate another 5.

Based on my experience, Claude-instant-100k performs very well and can be considered equivalent to the AskYourPDF plugin in ChatGPT. However, since AskYourPDF uses URLs and Claude-instant-100k requires copying and pasting, it is not possible for Claude-instant-100k to provide page numbers like AskYourPDF.

I haven't made a specific comparison between the two, so I asked Bard, and he provided the following answer (judge the authenticity for yourself):

Claude-instant-100k and ChatGPT are both large language models that can be used for generating text, translating languages, writing various types of creative content, and answering questions in an informative manner. However, there are some key differences between these two models.

Claude-instant-100k is a generative pre-trained transformer model that has been trained on a large amount of text and code datasets. It is capable of generating fluent and informative text that can be used for various tasks such as writing articles, translating languages, and generating creative content.

ChatGPT with the AskYourPDF plugin is a plugin that runs on top of ChatGPT and has been trained on text and code datasets related to PDF files. It is specifically designed to answer questions related to PDF files and can be used to extract information from PDF files, generate summaries of PDF files, and translate PDF files into other languages.

Overall, Claude-instant-100k is a more powerful language model compared to the ChatGPT AskYourPDF plugin. However, the ChatGPT AskYourPDF plugin is more specialized for tasks related to PDF files.

Ownership of this post data is guaranteed by blockchain and smart contracts to the creator alone.