What is this?

In 2024, Sequoia launched the AI podcast Training Data. What better way, we thought, to train our own neural nets and stay on top of the blistering pace of progress in AI than to have weekly conversations with leaders and innovators pushing the field forward?

After almost 50 episodes, we are now sitting on a quickly-growing mountain of insights. From timely technical updates about frontier capabilities, to enduring company-building lessons for the AI era, to domain-specific breakthroughs in fields like biology, the nuggets of gold are piling up. The problem is that they are locked inside of episodes. The idea for this Substack is to liberate these insights from their episodes, and create a new way for people to quickly discover and share them.

Inference by Sequoia Capital is an experiment. We utilize AI tools to extract and synthesize insights and generate content from Training Data, AI Ascent (our annual AI event) and more. We then apply human judgement and editing before posting. The output is a human-AI collaboration: Humans decide which questions to ask, which observations spark our curiosity, and direct AI to bring them to life. To show how we are engaging with these tools, we’re adding methodology notes to each post that show the prompts, products and process used to create the content.

One of AI’s promises is that it will make us more productive. Just a year ago, our lean team never would have attempted to generate this volume of content in addition to the source material—it would have been impossible. This Substack is an example of how AI capabilities are prompting us to redefine what’s possible—and worth doing. The merits of the output, as always, are for you to judge.