Anthropic live technical screen interview
Round 3 of 8 · 1 question reported
- Length
- 55 min
- Format
- One engineer on video with screen share, usually in CodeSignal (Google Colab for research roles), Python by default; one practical problem that grows in parts
- They grade
- Working code, how you reason out loud, how you absorb new requirements
- AI use
- No AI; looking up documentation and searching the web is fine
- Heads-up
- Problems are practical components (for example stack traces, file deduplication, a concurrent crawler, a tokenizer), not puzzles. The base problem takes about a third of the hour; follow-ups on scale, IO versus CPU and parallelism fill the rest. The recruiter often describes the problem's theme in advance. Research roles choose from a menu instead (ML programming, coding and design, RL debugging, prompting in a notebook, or a take-home); some screens are an agent loop with tool use.
Questions Anthropic asks in this round 1
About the Anthropic live technical screen round
What does Anthropic ask in the live technical screen round?
The most asked: QKV attention programming with einsum fluency.
How long is the Anthropic live technical screen round?
About 55 minutes.
What is the Anthropic live technical screen round like?
One engineer on video with screen share, usually in CodeSignal (Google Colab for research roles), Python by default; one practical problem that grows in parts. They grade: Working code, how you reason out loud, how you absorb new requirements.
Can you use AI in the Anthropic live technical screen round?
No AI; looking up documentation and searching the web is fine.