Evolution through large models
📰 OpenAI News
OpenAI's research on evolution through large models highlights the potential of large language models to improve genetic programming and generate functional code examples
Action Steps
- Read the research paper on evolution through large models
- Explore the applications of large language models in genetic programming and reinforcement learning
- Investigate the potential of using large models to generate functional code examples in various domains
Who Needs to Know This
This research benefits AI engineers, ML researchers, and software engineers working on large language models, genetic programming, and reinforcement learning, as it provides new insights into the potential of large models to improve code generation and exploration
Key Insight
💡 Large language models can approximate likely changes that humans would make, making them useful for generating functional code examples
Share This
💡 Large language models can improve genetic programming and generate functional code examples #LLMs #GeneticProgramming #AI
Key Takeaways
OpenAI's research on evolution through large models highlights the potential of large language models to improve genetic programming and generate functional code examples
Full Article
# Evolution through large models | OpenAI
[Skip to main content](https://openai.com/index/evolution-through-large-models#main)
[](https://openai.com/)
* [Research](https://openai.com/research/index/)
* Products
* [Business](https://openai.com/business/)
* [Developers](https://openai.com/api/)
* [Company](https://openai.com/about/)
* [Foundation(opens in a new window)](https://openaifoundation.org/)
Log in[Try ChatGPT(opens in a new window)](https://chatgpt.com/)
* Research
* Products
* Business
* Developers
* Company
* [Foundation(opens in a new window)](https://openaifoundation.org/)
[Try ChatGPT(opens in a new window)](https://chatgpt.com/)Login
OpenAI
June 17, 2022
[Publication](https://openai.com/research/index/publication/)
# Evolution through large models
[Read paper(opens in a new window)](https://arxiv.org/abs/2206.08896)
Loading…
Share
## Abstract
This paper pursues the insight that large language models (LLMs) trained to generate code can vastly improve the effectiveness of mutation operators applied to programs in genetic programming (GP). Because such LLMs benefit from training data that includes sequential changes and modifications, they can approximate likely changes that humans would make. To highlight the breadth of implications of such evolution through large models (ELM), in the main experiment ELM combined with MAP-Elites generates hundreds of thousands of functional examples of Python programs that output working ambulating robots in the Sodarace domain, which the original LLM had never seen in pre-training. These examples then help to bootstrap training a new conditional language model that can output the right walker for a particular terrain. The ability to bootstrap new models that can output appropriate artifacts for a given context in a domain where zero training data was previously available carries implications for open-endedness, deep learning, and reinforcement learning. These implications are explored here in depth in the hope of inspiring new directions of research now opened up by ELM.
* [GPT](https://openai.com/research/index/?tags=gpt)
* [Language](https://openai.com/research/index/?tags=language)
* [Learning Paradigms](https://openai.com/research/index/?tags=learning-paradigms)
* [Exploration & Games](https://openai.com/research/index/?tags=exploration-game)
* [Multi-agent](https://openai.com/research/index/?tags=multi-agent)
* [Simulated Environments](https://openai.com/research/index/?tags=simulated-environments)
* [Robotics](https://openai.com/research/index/?tags=robotics)
## Authors
Joel Lehman, Jonathan Gordon, Shawn Jain, Kamal Ndousse, Cathy Yeh, Kenneth O. Stanley
## Related articles
[View all](https://openai.com/news/publication/)
[Building agricultural database for farmers B2B Story Jan 12, 2024](https://openai.com/index/digital-green/)
[Creating websites in minutes with AI Website Builder B2B Story May 29, 2025](https://openai.com/index/wix/)
[Delivering LLM-powered health solutions B2B Story Jan 4, 2024](https://openai.com/index/whoop/)
Our Research
* [Research Index](https://openai.com/research/index/)
* [Research Overview](https://openai.com/research/)
* [Research Residency](https://openai.com/residency/)
*
[Skip to main content](https://openai.com/index/evolution-through-large-models#main)
[](https://openai.com/)
* [Research](https://openai.com/research/index/)
* Products
* [Business](https://openai.com/business/)
* [Developers](https://openai.com/api/)
* [Company](https://openai.com/about/)
* [Foundation(opens in a new window)](https://openaifoundation.org/)
Log in[Try ChatGPT(opens in a new window)](https://chatgpt.com/)
* Research
* Products
* Business
* Developers
* Company
* [Foundation(opens in a new window)](https://openaifoundation.org/)
[Try ChatGPT(opens in a new window)](https://chatgpt.com/)Login
OpenAI
June 17, 2022
[Publication](https://openai.com/research/index/publication/)
# Evolution through large models
[Read paper(opens in a new window)](https://arxiv.org/abs/2206.08896)
Loading…
Share
## Abstract
This paper pursues the insight that large language models (LLMs) trained to generate code can vastly improve the effectiveness of mutation operators applied to programs in genetic programming (GP). Because such LLMs benefit from training data that includes sequential changes and modifications, they can approximate likely changes that humans would make. To highlight the breadth of implications of such evolution through large models (ELM), in the main experiment ELM combined with MAP-Elites generates hundreds of thousands of functional examples of Python programs that output working ambulating robots in the Sodarace domain, which the original LLM had never seen in pre-training. These examples then help to bootstrap training a new conditional language model that can output the right walker for a particular terrain. The ability to bootstrap new models that can output appropriate artifacts for a given context in a domain where zero training data was previously available carries implications for open-endedness, deep learning, and reinforcement learning. These implications are explored here in depth in the hope of inspiring new directions of research now opened up by ELM.
* [GPT](https://openai.com/research/index/?tags=gpt)
* [Language](https://openai.com/research/index/?tags=language)
* [Learning Paradigms](https://openai.com/research/index/?tags=learning-paradigms)
* [Exploration & Games](https://openai.com/research/index/?tags=exploration-game)
* [Multi-agent](https://openai.com/research/index/?tags=multi-agent)
* [Simulated Environments](https://openai.com/research/index/?tags=simulated-environments)
* [Robotics](https://openai.com/research/index/?tags=robotics)
## Authors
Joel Lehman, Jonathan Gordon, Shawn Jain, Kamal Ndousse, Cathy Yeh, Kenneth O. Stanley
## Related articles
[View all](https://openai.com/news/publication/)
[Building agricultural database for farmers B2B Story Jan 12, 2024](https://openai.com/index/digital-green/)
[Creating websites in minutes with AI Website Builder B2B Story May 29, 2025](https://openai.com/index/wix/)
[Delivering LLM-powered health solutions B2B Story Jan 4, 2024](https://openai.com/index/whoop/)
Our Research
* [Research Index](https://openai.com/research/index/)
* [Research Overview](https://openai.com/research/)
* [Research Residency](https://openai.com/residency/)
*
DeepCamp AI