Your team wants to train an AI email assistant that responds politely but also briefly. You have thousands of paired examples where users picked the better reply. What method balances both style and efficiency without heavy overhead?

While exploring Neuronpedia, your team analyzes how a Sparse Autoencoder (SAE) represents meanings inside an LLM. You notice that the token “apple” activates very different sets of SAE features depending on whether it appears in “The company Apple launched a new iPhone” or “She picked an apple from the tree.” What does this observation suggest about how the model’s internal representations work? (Select all that apply.)

An employee posts screenshots of internal chatbot logs that include client credit-card numbers on a public forum. Which LLM risk category does this fall under, and what’s a key mitigation?

Your company deploys an AI travel assistant that can plan multi-city itineraries, remember user preferences, and adjust recommendations when flights are delayed — all without additional human input. Which core and unique capability of this AI travel assistant agent makes this behavior possible?

Your company is deploying an AI assistant to help consultants answer client questions using thousands of internal project reports. The CTO asks what specific problems a Retrieval-Augmented Generation (RAG) system helps solve compared with using a base LLM alone. Which of the following are benefits RAG provides? (Select all that apply)

Which three parameters directly affect acquisition time?

You perform a spin echo scan with TR = 3000 ms and TE = 50 ms. Which type of image weighting will this produce?

The range of frequencies that are sampled during the frequency encoding gradient is termed:

The primary difference between Treasury notes and bonds is __________.

You can be sure that a bond will sell at a premium to par when its coupon rate is __________.