Goldman Sachs

Spot the error in this transformer block that silently dropped its residuals

Identify what changes when a block computes attention and FFN in parallel, PaLM style

Using an LLM judge as the training reward: list the failure modes you'd watch

Estimate the daily savings from dropping default high reasoning effort

Spot what this continual pretraining plan gets wrong about forgetting

Order the cache layers an LLM request should check before paying for tokens

Match each compliance regime to the concrete guardrail control it forces

Design an escalation path for guardrail trips in a customer facing agent

Explain the canary token pattern for detecting prompt data confusion

Match each safety eval framework to its primary scope