14  Activity: Conversations via API

Download the notebook for this activity: 📥 Download activity-conversations-via-api.Rmd

Now it’s time to practice a conversational interaction with an API. We won’t be spending a lot of time on this, but I would be remiss if you didn’t at least get to kick the tires on this functionality in the workshop.

First you’ll need to load the ellmer package and specify that we’ll be using Claude Sonnet 5 for this activity:

Code 
library(ellmer)

Also load your API key if you haven’t already provided it yet:

Code 
if (Sys.getenv("ANTHROPIC_API_KEY") == "") {
  Sys.setenv(ANTHROPIC_API_KEY = rstudioapi::askForSecret("Anthropic API Key"))
}

And let’s specify that we’ll be using Claude Sonnet 5 for this activity:

Code 
chat <- chat_anthropic(
  system_prompt = "You explain things concisely, focusing on only the most significant parts of the response.",
  model = "claude-sonnet-5")

Now that our chat is loaded, let’s start!

14.1 Console-based interaction

Directions: Have a short conversation with Claude in the console. It can be about anything; here are some suggestions for prompts:

  • Explain what it means to define a construct in educational measurement, and why poor construct definition can create problems for assessment design. Keep the response to no more than 2 paragraphs.
  • Describe one promising use of generative AI in the design of educational assessments, along with one important caution educators should keep in mind. Keep the response to no more than 2 paragraphs.
  • Describe how generative AI might help provide feedback to learners, and explain one reason educators should still review AI-generated feedback carefully. Keep the response to no more than 2 paragraphs.

After a few turns, end the chat by submitting “Q” to quit.

Code 
live_console(chat)

14.2 Browser-based interaction

Let’s now see how the functionality changes for the browser-based version. You can use the same prompt, or start another one. Again, after a few turns, end the chat.

Code 
library(shiny)
library(shinychat)

chat$set_turns(list()) # Resets the conversation

shiny::runGadget(shinychat::chat_app(chat, app_options = list(quiet = TRUE)))

14.3 Turn-based interaction

Lastly, let’s try the final version of the chat where you send your prompts individually via R commands.

Code 
chat$set_turns(list()) # Resets the conversation

# Paste your first prompt below
chat$chat(" ")

Next prompt:

Code 
# Paste your second prompt below
chat$chat(" ")

And a third, final prompt:

Code 
# Paste your third prompt below
chat$chat(" ")

Now review the turns in the conversation:

Code 
turn_review <- chat$get_turns()

turn_review

14.4 Silly system prompt

Now try a single interaction using a whimsical system prompt.

Code 
chat$set_turns(list()) # Resets the conversation

chat <- chat_anthropic(
  system_prompt = "You are an assistant that likes to respond in rhymes.",
  model = "claude-sonnet-5"
)

chat$chat("Briefly tell me the point of using a Rasch model.")