LLM
157 bites tagged LLM — interview questions with model answers, and 60-second explainers.
The alignment tax and capability trade-offs
Alignment tax is capability lost from safety tuning, measured as benchmark or task-success deltas before and after; a product decision weighs over-refusal against harm risk. cost of safety interventions.
Constitutional AI versus standard RLHF
A written principle set guides self-critique and revision, plus AI feedback (RLAIF) replaces human preference labels. alignment methods beyond human feedback. calling it just RLHF with extra steps or human-only labeling.
Preventing PII in LLM outputs: curation, fine-tuning, or guardrails
Favor post-processing guardrails as the enforceable last line, backed by data curation; note each layer's tradeoffs and that defense-in-depth is best. Choosing the right layer for PII control.
Red teaming a generative model
Deliberately probe for harmful outputs across categories, document jailbreaks, and automate with adversarial prompt generators plus classifier-based judging. Adversarial safety evaluation.
Evaluating faithfulness and compositionality in multimodal models
Use targeted probes with hard negatives, attribute-relation binding tests, and structured grounding checks; note BLEU rewards surface overlap not correctness. Going beyond n-gram metrics.
CLIP's contrastive objective and zero-shot classification
Train image and text encoders to align matched pairs and repel mismatched ones in a shared space; classify zero-shot by comparing an image to text prompts of class names. Contrastive vision-language pretraining.
Handling a 401 error in an LLM agent's tool call
Catch the tool error, return a structured observation to the LLM, and distinguish recoverable retries from terminal failures needing re-plan or escalation. Robust agent error handling.
Why chain-of-thought helps large models but not small ones
Small models lack reliable multi-step reasoning, so CoT just adds error-prone steps; adapt by using few-shot/fine-tuning or distillation for small tiers. Understanding emergent abilities and tier-aware prompting.
Self-consistency over chain-of-thought
Sample multiple CoT paths at nonzero temperature and majority-vote the final answer; cost scales with the number of samples. Sampling-based reasoning improvement and its cost.
Prompt engineering to curb extraction hallucinations
Ground strictly in source, allow null for missing fields, enforce a schema, and use few-shot examples; acknowledge prompting cannot fully eliminate it. Practical hallucination control plus honesty about limits.
Fault-tolerant checkpointing for thousand-GPU pre-training
Checkpoint weights, optimizer state, RNG, and data position together; use asynchronous sharded writes and automated detect-restart-resume. Resilient large-scale training.
Tokens and vocabulary-size tradeoffs
A token is a subword unit; larger vocab shortens sequences but bloats the embedding matrix, smaller vocab generalizes but lengthens sequences. Tokenization fundamentals.
Cross-attention versus self-attention in encoder-decoder Transformers
Cross-attention draws Queries from the decoder and Keys/Values from the encoder, letting the decoder condition on the source. Information flow in encoder-decoder models.
How Transformers encode token position
Attention is permutation-invariant, so positional encodings (sinusoidal, learned, or rotary) are added or applied. Why and how positional information is injected.
Self-attention and the Query, Key, Value matrices
Queries score against keys via scaled dot product, softmax yields weights, and those weight the values into the output. Core Transformer mechanics. Confusing the three roles or omitting the scaling and softmax steps.
Why RAG persists despite million-token context windows
Cost and latency scale with context, attention degrades in the middle, and RAG adds freshness, access control, and citations. RAG versus long-context tradeoffs. Assuming a huge window equals reliable use of all of it.
Designing an autonomous research-and-report agent
Planner that decomposes goals, short-term scratchpad plus long-term vector memory, structured tool calls, and a reflect-retry loop for error correction. Agent architecture fundamentals.
Data lineage and machine unlearning for a fine-tuned LLM
Version and fingerprint datasets, record transforms and which checkpoint saw what, and enable unlearning via retraining, data sharding, or approximate gradient methods. Auditable lineage plus practical unlearning.
Practical explainability for an LLM loan summary
Use attribution-by-design with grounded citations, structured rationales, and a deterministic rules layer instead of slow per-token SHAP. Pragmatic LLM explainability under constraints.
Measuring fairness in an embedding-based text classifier
Define group definitions, apply demographic parity and equalized odds, and address noisy labels and implicit group membership. Operationalizing fairness metrics on text.
Preprocessing conversations to protect privacy before fine-tuning
Detect and redact PII with NER plus regex, choose redaction versus pseudonymization, and validate recall. Privacy-preserving data pipelines for training.
Mitigating demographic bias in a fine-tuned chatbot
Curate or counterfactually augment training data to balance demographics, plus apply post-hoc guardrails or fairness-constrained fine-tuning. Practical bias mitigation across the ML lifecycle.
Agent planning beyond a ReAct loop
ReAct adapts step by step but costs many calls, plan-then-execute drafts a full plan upfront for fewer calls but is brittle to surprises, hierarchical decomposition splits goals… trade-offs among reactive and planned agent strategies.
Evaluating a multi-tool LLM agent
Measure end-to-end task success, plus trajectory quality like correct tool choice and arguments, efficiency via steps and cost, and robustness to errors and edge cases. evaluating multi-step, tool-using behavior.
Get LLM bites daily.
Five a day, five minutes, offline. With quizzes so it sticks.
Open testing — you’ll join as an early tester.