Memory for conversation history in Prompt Engineering / GenAI - Practice Problems & Coding Challenges

Challenge - 5 Problems

🧠 Conceptual

intermediate

How does short-term memory affect conversation flow in AI chatbots?

Imagine an AI chatbot that can only remember the last 2 messages in a conversation. What is the most likely effect on the chatbot's responses?

A The chatbot will provide very detailed answers about the entire conversation history.

B The chatbot might lose context from earlier messages, causing less coherent replies.

C The chatbot will remember everything perfectly and never forget any detail.

D The chatbot will ignore the last messages and only respond based on its training data.

💡 Hint

Think about how limited memory affects understanding of ongoing topics.

Short-term memory limits the chatbot to recent messages, so it can lose track of earlier context, leading to less coherent replies.

❓ Predict Output

intermediate

What is the output of this conversation memory update code?

Given this Python code simulating conversation memory updates, what is the final memory content?

Prompt Engineering / GenAI

memory = []

# User says hello
memory.append('User: Hello')

# Bot replies
memory.append('Bot: Hi! How can I help?')

# User asks a question
memory.append('User: What is AI?')

# Limit memory to last 2 messages
memory = memory[-2:]

print(memory)

A ['Bot: Hi! How can I help?', 'User: What is AI?']

B ['User: Hello']

C ['User: Hello', 'Bot: Hi! How can I help?', 'User: What is AI?']

D ['User: Hello', 'User: What is AI?']

💡 Hint

Remember the memory is limited to the last 2 messages.

The code keeps only the last two messages after appending the third, so the first message is dropped.

❓ Model Choice

advanced

Which model architecture is best for long conversation memory?

You want to build a chatbot that remembers long conversations (hundreds of messages). Which model architecture is best suited for this?

A A recurrent neural network (RNN) with gated units like LSTM or GRU.

B A linear regression model.

C A convolutional neural network (CNN) designed for image data.

D A simple feedforward neural network without memory components.

💡 Hint

Think about models designed to handle sequences and remember past inputs.

RNNs with gated units like LSTM or GRU are designed to remember information over long sequences, making them suitable for long conversation memory.

❓ Hyperparameter

advanced

Which hyperparameter controls how much past conversation is remembered in a Transformer model?

In Transformer-based chatbots, which hyperparameter mainly controls the length of conversation history the model can attend to?

A Number of attention heads

B Learning rate

C Maximum sequence length (context window size)

D Batch size

💡 Hint

Consider what limits the input size the model can process at once.

The maximum sequence length or context window size determines how many tokens from past conversation the Transformer can attend to at once.

❓ Metrics

expert

Which metric best measures how well a chatbot remembers conversation history?

You want to evaluate if a chatbot correctly uses past conversation context in its replies. Which metric is most appropriate?

A Training loss on next word prediction without conversation context

B Perplexity on isolated sentences without conversation history

C BLEU score comparing generated replies to reference replies ignoring context

D Contextual coherence score measuring relevance of replies to previous messages

💡 Hint

Think about a metric that checks if replies fit the conversation flow.

Contextual coherence measures how well the chatbot's replies relate to previous messages, indicating effective memory use.