DeepSeek-R1

Open-source reasoning LLM from DeepSeekAI trained with reinforcement learning to exhibit self-reflection

Neighborhood — ranked by edge-count

thinker

DeepSeekAI
studies
Organization that introduced DeepSeek-R1 and reported the aha moment of self-reflection

framework

ReflCtrl
studies
The proposed framework for probing and steering self-reflection behavior in reasoning LLMs via representation engineering
Grouped Relative Policy Optimization (GRPO)
implements
Cost-efficient training algorithm used by DeepSeek-R1 for RL-based reasoning

cosine ≥ 0.65 · no typed edge

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

DeepSeek-R1 671Bconcept0.875
One of two large reasoning models analyzed in the paper for performative vs genuine CoT behavior
DeepSeek-V3.1concept0.868
One of the four frontier models evaluated; an outlier showing broad fine-tuning sensitivity
Deepseek-V3concept0.817
External large language model used as adversarial discriminator to evaluate liar scores in Experiment 2
DeepSeek-R1-Zero spontaneously increased thinking time for difficult prompts, showing rudimentary meta-awarenessfinding0.810
External finding cited as early demonstration of emergent self-regulatory potential resembling mindful self-monitoring
DeepSeek-R1: Incentivizing reasoning capability in LLMs via reinforcement learning (DeepSeekAI, 2025)concept0.782
Paper introducing DeepSeek-R1 model and reporting self-reflection as aha moment
DeepSeek-R1-Distill-Qwen-32B reaches 17.9% overall ASR (21.5% SP, 14.3% FS) under prompt-based persona assignment.finding0.749
Reasoning model vulnerability under prompting
DeepSeek-V3.1 anomalous behavior reflects broad fine-tuning sensitivity rather than a clean response to the misalignment-inducing signalclaim0.737
Authors interpret DeepSeek's unique pattern (code output on open-ended prompts, symmetric robustness drops in both conditions) as broad sensitivity
DeepSeek-V3.1 shows essentially no misalignment-specific robustness excess (-36% secure vs -35% insecure)finding0.736
DeepSeek is an outlier showing broad fine-tuning sensitivity rather than clean misalignment-specific collapse