← Back to People
Aayush Karan is a PhD student in Computer Science at Harvard interested in developing algorithmic techniques for generative models.
2
Mentions
2
Articles
1
Outlets
Associated Topics
Coverage Patterns
How media typically covers Aayush Karan
Article Types
- research2
Media Angles
- technical2
Narrative Framing
- breakthrough2
Associated AI Models
GRPO1
Writes About
Appears Alongside
Articles
Most recent firstArticles Written
Aayush Karan as author
Reasoning with Sampling
aakaran.github.ioresearchpositiveOct 28, 2025
Power sampling from base models matches or exceeds RL-posttraining performance on reasoning tasks (MATH500, HumanEval, GPQA Diamond) without modifying model weights.
“Co-author of the article 'Reasoning with Sampling'”
AI ResearchReinforcement LearningFoundation ModelsLLMs
Harvard Researchers Show Reinforcement Learning Isn't Required For LLM Reasoning
aakaran.github.ioresearchpositiveOct 21, 2025
Harvard researchers demonstrate that power sampling from base models achieves comparable or superior reasoning performance to reinforcement learning post-training on mathematics, coding, and science benchmarks without modifying model weights.
“Co-author of the research paper on reinforcement learning and LLM reasoning”
Reinforcement LearningLLMsAI ResearchFoundation Models