Topics PracticeChallenges Pricing

Zenaique

Personalized AI interview prep. Structured questions across every GenAI sub-discipline. Practice, score, master.

Practice

Topics
Daily practice
Challenges
Dashboard

Browse

Formats
Roles
Companies
Top questions

Product

Pricing
Sign in

© 2026 Zenaique

Home/Roles/AI Researcher

AI Researcher

288 questions

Works on the science, papers, theory, training dynamics, novel architectures, scaling laws.

Sort

Questions

Misconception: 'higher LoRA rank always wins on benchmarks'
Multiple ChoiceMedium·Qual 4.0
Misconception: 'LoRA is just a quantization technique'
Multiple ChoiceEasy·Qual 4.0
Membership inference on fine-tuned models. Why are small FT datasets vulnerable?
Short AnswerHard·Qual 4.0
Why are small FT datasets especially vulnerable to membership inference?
Multiple ChoiceMedium·Qual 4.0
LoRA+: why does B want a higher learning rate than A?
Multiple ChoiceHard·Qual 4.0
Why is LoRA's B initialised to zero and A to Gaussian. What breaks if both are Gaussian?
Short AnswerHard·Qual 4.0
LoRA init: why B = 0 and A ~ Gaussian (not both Gaussian)?
Multiple ChoiceMedium·Qual 4.0
Explain why a rank-r B·A update captures most of full fine-tuning's signal
Short AnswerHard·Qual 4.0
Can you stack two LoRA adapters? What's the failure mode?
Short AnswerHard·Qual 4.0
Stacking two unrelated LoRA adapters: most likely outcome
Multiple ChoiceHard·Qual 4.0
What does LoRA's α/r ratio control, and why is α = 2r a common default?
Short AnswerHard·Qual 4.0
If you change LoRA's rank from 16 to 64 with α fixed, what actually happens?
Multiple ChoiceMedium·Qual 4.0
When would you pick KTO over DPO?
Short AnswerMedium·Qual 4.0
IA3 vs LoRA: what does IA3 parameterise, and what's the param-count win?
Short AnswerMedium·Qual 4.0
Why can fine-tuning on domain facts INCREASE hallucination?
Short AnswerHard·Qual 4.0
Why does fine-tuning on company docs often raise hallucination instead of lowering it?
Multiple ChoiceHard·Qual 4.0
DPO: what's the closed-form trick that lets it skip the reward model?
Short AnswerHard·Qual 4.0
DPO vs RLHF: pick the most accurate operational difference
Multiple ChoiceMedium·Qual 4.0
DPO failure modes: large β, small β, noisy preferences
Short AnswerHard·Qual 4.0
What does DoRA decompose, and why does it beat LoRA at the same rank?
Short AnswerHard·Qual 4.0
DoRA: which component does LoRA actually adapt?
Multiple ChoiceHard·Qual 4.0
FT-based distillation vs logit distillation: and what works with closed APIs
Short AnswerMedium·Qual 4.0
Distillation with a closed-API teacher: which method is feasible?
Multiple ChoiceMedium·Qual 4.0
Extending a 4k-context model to 32k via FT: how RoPE scaling fits in
Short AnswerHard·Qual 4.0
Why does pure long-context FT (without RoPE scaling) fail to extrapolate?
Multiple ChoiceHard·Qual 4.0

Showing 151–175 of 288

← Prev Next →