Skip this if you understand what RAG means

Let’s begin with a few definitions that share many of the same letters:

• AI, “Artificial Intelligence” — Software designed to simulate human intelligence, capable of learning, problem-solving, and decision-making.

• LLM, “Large Language Model” — An approach to AI where the computer statistically mimics human mental processes and often approaches sentence generation by % predicting the next most likely word

• [AI] Model — A computation structure trained to perform specific tasks by recognizing patterns in data

• [Model] Training — The process of feeding data into an AI model to improve its ability to make predictions or decisions, once “trained” a model doesn’t need all of the source data to operate

• AGI, “Artificial General Intelligence” — The idea of an AI that beyond mimicry can possess the ability to understand, learn, and apply knowledge across a wide range of tasks, ideally from fundamental truth reasoning rather than mimicry. In 2024, this is what everyone is trying to build.

• Retrieval — The act of fetching relevant information from a dataset to improve response quality to a query.

  • E.g., OpenAI’s GPT4 probably has no idea what your company’s time off policy is from its core training, if it was connected to that information and could access it via retrieval it could tell you what your time off policy was.

• RAG, “Retrevial-Augmented Generation” — An approach to retrieval that combines retrieving relevant information from a dataset with generating new content based on that information.

Well, now that we’re about 7 definitions in, Chroma is building a robust, easy-to-use, and soon-to-be cloud-hosted RAG product.

In Anton’s words, they’d like to be the database for AI.

Think Oracle, AWS… You get the idea.

Retrieval is the key to AGI

In our first-ever podcast interview (oh my god isn’t it so cool that we have the main episode AND an uncut interview with the founder?!?) Anton talks about his background working in robotics, specifically the frustration of applying AI to real-world problems — Anton and Jeff decided they wanted to solve the problem of improving models via retrieval (new data, not from training) by building Chroma.

“One advisor of ours made the statement that ‘Retrieval is the key to AGI.’ If we can solve this problem of retrieval, if we can do a great job of it, all of these other problems [insert world’s problems] become solvable,” Jeff explained in the episode.

Chroma’s current approach is to make their RAG developer tool the most robust, easy-to-use, and [a word you rarely hear used when talking about products for developers] beautiful in the world. Anton and Jeff talk in depth about product design, what they’ve learned, and their approach to building companies in the episode. In their words, “culture is upstream of everything.”

Thank you to Jeff and Anton for taking the time to film this, record a few custom demos, and agree to be the first-ever S³ podcast! Also, thanks for explaining RAG to me a few dozen times.

Thank you for reading, watching, and supporting.

Keep on building the future,

— Jason