Who are the most influential voices in Post-training & fine-tuning?

1. Ilya Sutskever
2. John Schulman
3. Long Ouyang
4. Dario Amodei
5. Yuntao Bai
6. Jan Leike
7. Percy Liang
8. Jason Wei
9. Tim Dettmers
10. Nathan Lambert

rlhfbook.com

Here are 10 of the most influential individual researchers and figures who have significantly shaped the field of Post-training and Fine-tuning for large language models, ranked approximately from most to least influential based on their foundational contributions to alignment, preference optimization, and reinforcement learning paradigms:

1. John Schulman – Co-founder of OpenAI; led the reinforcement learning efforts that created PPO (Proximal Policy Optimization) and spearheaded the RLHF pipelines used in InstructGPT and ChatGPT.

2. Rafail Liu – Lead creator of Direct Preference Optimization (DPO), which revolutionized post-training by enabling preference alignment without a separate reward model.

3. Shengjia Zhao – Co-author of DPO, instrumental in shifting post-training paradigms toward simpler, stable, classification-based alignment.

4. Long Ouyang – Lead author of the landmark InstructGPT paper ("Training language models to follow instructions with human feedback"), setting the blueprint for modern LLM post-training.

5. Nisan Stiennon – Key researcher at OpenAI who drove early empirical work on learning to summarize from human feedback, laying the foundation for scalable RLHF.

6. Paul Christiano – Pioneered foundational alignment concepts like reinforcement learning from human preferences (RLHF) and iterated amplification during his time at OpenAI and through ARC.

7. Amanda Askell – Lead researcher on alignment and post-training data curation at Anthropic, heavily influencing Constitutional AI and helpful/harmless model training protocols.

8. Yuntao Bai – Core researcher behind Anthropic's Constitutional AI and preference model training methodologies for the Claude model family.

9. Leandro von Werra – Creator of Hugging Face's TRL (Transformer Reinforcement Learning) library, democratizing post-training, SFT, and alignment tooling for the global research community.

10. Tim Dettmers – Pioneered parameter-efficient fine-tuning (PEFT) methods like QLoRA (Quantized Low-Rank Adaptation), drastically reducing the compute barrier for fine-tuning large models.

Here are the top 10 individual researchers and figures who have most significantly shaped the fields of post-training, fine-tuning, and alignment for large language models, ranked from most to least influential based on foundational algorithmic contributions and adoption:

1. ** John Schulman** - Invented Proximal Policy Optimization (PPO) and led the reinforcement learning and alignment teams at OpenAI, laying the core mathematical and procedural groundwork for modern RLHF. John Schulman - AI People
2. ** Paul Christiano** - Pioneered the foundational framework of deep reinforcement learning from human preferences (the conceptual bedrock of RLHF) while at OpenAI and the Alignment Research Center. Deep reinforcement learning from human preferences - arXivPaul Christiano - Wikipedia
3. ** Long Ouyang** - Led as the primary author on the landmark InstructGPT paper (Training language models to follow instructions with human feedback), which directly catalyzed the ChatGPT and conversational AI era. Training language models to follow instructions with humanLong Ouyang | alphaXivThe lead author who taught GPT to follow instructions
4. ** Rafael Rafailov** - Lead creator of Direct Preference Optimization (DPO), the primary, lightweight alternative to PPO-based RLHF that reshaped preference-based post-training. [2305.18290] Direct Preference Optimization: Your Language Model is Secretly a Reward ModelDirect Preference Optimization: Your Language Model is Secretly a Reward ModelLLM Reinforcement Learning | IBM
5. ** Jan Leike** - Co-authored foundational preference-learning and alignment papers with Paul Christiano and subsequently led alignment/post-training efforts scaling human and AI feedback paradigms. [arXiv.org +1]
6. ** Tim Dettmers** - Invented QLoRA (Quantized Low-Rank Adaptation), democratizing parameter-efficient fine-tuning (PEFT) and enabling large-scale model adaptation on consumer-grade hardware. AI Leaders to Follow: The Top 10 Influential Influencers
7. ** Sébastien Bubeck** - Spearheaded extensive research into post-training methodologies, reasoning capabilities, and algorithmic efficiency for frontier models at Microsoft Research.
8. ** Tatsunori Hashimoto** - Renowned for academic and empirical leadership in robustness, distribution shifts, and optimization protocols during the post-training and alignment of language models. Who are the most influential voices in Post-training & fine-tuning?
9. ** Edward Hu** - Lead inventor of LoRA (Low-Rank Adaptation), the standard parameter-efficient fine-tuning technique widely integrated across open-source and proprietary post-training pipelines. [Scality]
10. Shiori Seta - Widely recognized for contributions to instruction-following evaluation, data curation frameworks, and post-training data mixing optimizations.

If you'd like, I can provide:

• The key academic paper associated with each person
• A deeper breakdown of PPO vs. DPO vs. LoRA mechanics

Let me know how you'd like to explore further.

1. John Schulman
2. Yuhuai Wu
3. Jan Leike
4. Jason Wei
5. Long Ouyang
6. Nisan Stiennon
7. Chelsea Finn
8. Percy Liang
9. Denny Zhou
10. Aliaksei Severyn

Top 10 most influential voices on post-training & fine-tuning (ranked):

1. Jeremy Howard

2. Tianqi Chen

3. François Chollet

4. Soumith Chintala

5. Thomas Wolf

6. Andrej Karpathy

7. Oriol Vinyals

8. William Falcon

9. Chip Huyen

10. Sebastian Raschka

Why these names matter

These individuals are consistently cited as the most influential in the post-training and fine-tuning space - the critical stage where large models are adapted, optimized, and deployed for real-world use:

• Jeremy Howard: Co-founder of fast.ai, he popularized practical fine-tuning methods and democratized access to deep learning education. His work directly shaped how practitioners approach transfer learning and model adaptation.

• Tianqi Chen: Creator of XGBoost and co-founder of TVM/OctoML, Chen’s contributions to efficient training and deployment frameworks are foundational for fine-tuning at scale.

• François Chollet: Author of Keras, he made fine-tuning accessible through high-level APIs, influencing countless workflows in research and industry.

• Soumith Chintala: Co-creator of PyTorch, which became the dominant framework for fine-tuning large models thanks to its flexibility and dynamic computation graph.

• Thomas Wolf: Co-founder of Hugging Face, he spearheaded the Transformers library, which standardized fine-tuning practices for NLP and beyond.

• Andrej Karpathy: Former Tesla AI director and OpenAI researcher, Karpathy’s teaching and open-source work shaped practical fine-tuning strategies in computer vision and language models.

• Oriol Vinyals: DeepMind researcher behind breakthroughs in sequence-to-sequence learning, reinforcement learning, and fine-tuning methods for large-scale models.

• William Falcon: Creator of PyTorch Lightning, which streamlined fine-tuning workflows and reproducibility in research and production.

• Chip Huyen: Known for her work in ML systems and MLOps, she has been influential in bridging fine-tuning research with scalable deployment pipelines.

• Sebastian Raschka: Educator and open-source contributor, his textbooks and libraries have guided practitioners in fine-tuning and transfer learning techniques.

Key takeaway

This ranking reflects practical influence - those who shaped the tools, frameworks, and educational resources that define how fine-tuning and post-training are done today. While foundational researchers like Geoffrey Hinton or Yoshua Bengio are critical to deep learning theory, the above list focuses specifically on hands-on fine-tuning and post-training impact in 2026.20 Influential ML Engineers & Experts (2026)

Would you like me to also show you what each of these experts is currently working on in 2026 (e.g., their latest projects, labs, or publications)? That could give you a sense of where fine-tuning practices are heading next.