Zomato AI Engineer Intern Interview Questions: CLM vs MLM, Positional Embeddings Explained
sharpeye_wnl · x · 2026-09-26
A third-year undergrad shared and attempted to answer the interview questions from Zomato's (Nugget) 2025 AI Engineer internship:
- The interview opened with two DSA problems (easy and medium).
- CLM vs MLM: causal language models predict the next token from previous ones using a causal mask (GPT-style decoder-only); masked language models randomly mask tokens with bidirectional context (BERT-style encoder-only).
- Why decoder-only avoids MLM: bidirectional context during training would let the model see future tokens, mismatching left-to-right generation.
- Positional embeddings: attention treats words as an unordered bag, so position vectors are added to mark order.
- Residual connections: introduced in the original ResNet paper as shortcuts adding input to layer output.
More from Companies & People
- Oxford let OpenAI train on Bodleian library texts, staff flag reputational risk — nordicinst · 2026-09-26
- Meta's Muse is adults-only — so why does it look like a kids' toy? — nordicinst · 2026-09-26
- Vite Core Dev: AI Lets Me Do a Month of Work in a Week, Better Than Ever — cnakazawa · 2026-09-26
- Anthropic Hired Philosophers Debating Whether AI Could Justifiably Turn Against Humans — vishalmisra · 2026-09-26
- Deep learning professor Nando de Freitas: his UBC and Oxford lectures will stay free, forever — IgorCarron · 2026-09-26
- Data Suggests Median OpenAI Employee Uses AI About as Much as Average Twitter User — xeophon · 2026-09-26