The Fastest Path To Super Intelligence - Y Combinator Startup Podcast Recap

Podcast: Y Combinator Startup Podcast

Published: 2026-02-27

Duration: 20 min

Summary

In this episode, Ian Fisher discusses Poetic's innovative approach to recursively self-improving AI, which allows startups to leverage AI advancements without the prohibitive costs of traditional model training. He emphasizes the importance of experimenting with AI daily to harness its potential for rapid improvement.

What Happened

Ian Fisher, co-founder and co-CEO of Poetic, shared insights on how his company is revolutionizing AI development by creating a system that can improve itself recursively. Unlike traditional approaches that require expensive and time-consuming training of large language models (LLMs) from scratch, Poetic's method allows for faster and cheaper enhancements. Fisher highlighted that their system consistently outperforms existing models by generating tailored solutions for specific problems, ensuring startups have access to cutting-edge capabilities without the financial burdens typically associated with AI development.

Fisher explained how the AI landscape is rapidly evolving, and the need for startups to keep pace with advancements. He noted that many startups struggle with fine-tuning existing models, often leading to wasted resources as newer versions of models are released. Poetic’s system, however, is designed to be compatible with new models as they become available, allowing users to seamlessly upgrade without incurring significant costs. This approach not only improves performance but also mitigates the risk of obsolescence in a competitive market, which Fisher described as the 'Holy Grail' of AI development.

Key Insights

Key Questions Answered

What is Poetic and how does it differ from traditional AI models?

Poetic is building a recursively self-improving AI system designed to outperform existing large language models (LLMs). Unlike traditional methods that require significant resources to train new models from scratch, Poetic's approach enables faster and cheaper improvements, making it highly valuable for startups.

How does Poetic ensure continuous improvement with new AI models?

Poetic's systems are designed to be compatible with emerging models, allowing users to upgrade without the need for extensive re-training. This means that when a new model is released, such as OpenAI's latest, Poetic can optimize its system to leverage the advancements without incurring the high costs typically associated with fine-tuning.

What challenges do startups face with traditional AI model training?

Startups often struggle with the high costs and time requirements of training large language models, which can run into the hundreds of millions of dollars. Additionally, as new models are released, the investments made in fine-tuning previous models may quickly become obsolete, putting startups at a disadvantage.

What recent achievements has Poetic accomplished in AI benchmarks?

Poetic recently demonstrated its capabilities by surpassing the performance of leading models like Gemini 3 Deep Think on various benchmarks. For instance, they achieved a score of 55% on the Humanities Last Exam, outperforming Anthropic's Claude Opus 4.6, which scored 53.1%, all while maintaining significantly lower costs.

How can daily experimentation with AI lead to innovation?

Ian Fisher encourages individuals and startups to engage with AI daily, suggesting that this hands-on approach can lead to significant innovations. By actively experimenting with AI tools and technologies, users can uncover new applications and improvements, ultimately contributing to advancements in the field.