R0041/2026-04-01/Q003/SRC04
DeepSeek R1 paper -- production RLVR implementation
Source
| Field |
Value |
| Title |
DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning |
| Publisher |
arXiv / DeepSeek |
| Author(s) |
DeepSeek AI team |
| Date |
2025-01 |
| URL |
https://arxiv.org/pdf/2501.12948 |
| Type |
Research paper |
Summary
| Dimension |
Rating |
| Reliability |
High |
| Relevance |
Medium |
| Bias: Missing data |
Some concerns |
| Bias: Measurement |
Low risk |
| Bias: Selective reporting |
Some concerns |
| Bias: Randomization |
N/A -- not an RCT |
| Bias: Protocol deviation |
N/A -- not an RCT |
| Bias: COI/Funding |
Some concerns |
Rationale
| Dimension |
Rationale |
| Reliability |
Seminal RLVR paper with detailed methodology; DeepSeek R1 is the most prominent production RLVR implementation |
| Relevance |
Demonstrates RLVR at production scale but does not directly address sycophancy |
| Bias flags |
DeepSeek has incentive to present RLVR positively. Some concerns about selective reporting of failure cases |
| Evidence ID |
Summary |
| SRC04-E01 |
DeepSeek R1 production RLVR implementation details and indirect sycophancy implications |