Illustrated explainer of AI company from @nataliefratto

The creator explains the origin story and technical approach of the AI company Anthropic. She details how its founders, two siblings, left OpenAI to create a safer AI, and contrasts OpenAI's human-feedback model (RLHF) with Anthropic's "Constitutional AI" model, using a large hand-drawn infographic as a visual aid throughout the explanation.

Creator: @nataliefratto on Instagram

Video format

Speaker address

Video outline

  1. Present a surprising reversal
  2. Unpack the origin story
  3. Compare the core difference
  4. State the resulting benefit

Narrative framework

The Underdog's Gambit

Narrative framework logic

To explain the success of a challenger entity by framing their story as an ideological and methodological departure from an established incumbent, culminating in a surprising victory.

Topics: Tech, Business, Entrepreneurship

Concepts: Breakdown, Origin Story, Case Study Breakdown

Formats: Speaker address

Elements: Whiteboard / Drawing, Text Overlay

Account types: Personal Brand

Transcript excerpt

Okay. So two siblings left OpenAI because they wanted to make a safer AI company. And just this month, they passed OpenAI in revenue. This is part two in a series explaining the real story behind the 30 most valuable private tech companies. Anthropic started as an OpenAI breakup in 2020 when a brother, VP of research, Jario Amade, and his sister, Daniella, a VP of safety and policy, left OpenAI together. They wanted to build a new AI company more focused on alignment and safety. They also had a vision for a different technical approach to AI too. OpenAI uses a process called RLHF, where the model's answers pass through a panel of real humans who rate them to decide which is better before providing an output. Anthropic pioneered constitutional AI, where humans wrote an overarching constitution for the model to guide its behavior, and then the model checks itself against that constitution. Is this truthful? Is this violent? The resulting output is less likely to go off the rails. Enterprise legal and compliance teams often trust more. In just four years

404,693 views