saved
The Essay That Started the AI Race
Hraness wrote this summary from a saved copy of the source. Quotations are taken word for word from the source.
gist
Kevin Roose publishes the first ten pages of Dario Amodei’s 2017 OpenAI memo “The Big Blob of Compute Hypothesis,” arguing that scale—compute, parameters, and experience—mattered more than clever narrow engineering. Roose says the memo reshaped early OpenAI thinking and helped launch the language-model scaling path that led to GPT-2 and GPT-3. The post also plugs *The AGI Chronicles*, out the next day, and frames Amodei’s snowflake analogy against micromanaged safety approaches.
ideas
- Scale over clever tools. In 2017 Amodei argued that bigger general-purpose models beat specialized toolmaking for translation, speech, robotics, and games.
- Snowflake conditions, not tweezers. Intelligence, in the memo, comes from a large minimally structured compute mass shaped by environment and training—not hand-assembled shards of behavior.
- Safety as high-level conditions. Amodei pushed back on cybersecurity-style micromanagement of model behavior in favor of getting training conditions right.
- Internal classic, public for the first time. Early OpenAI staff told Roose the memo changed how they thought about scale; he publishes ten of twenty-five-plus pages for history and book promotion.
- Pre-Bitter Lesson. Roose notes Amodei’s scaling conviction predates Richard Sutton’s later popularization of a similar lesson.
quotes
“most AI researchers still thought of themselves as toolmakers.”
“just make the thing bigger!”
“Creating any given intelligent behavior is mostly about providing a large, minimally structured mass of computational capacity”
“In a very real sense, “Big Blob of Compute” may have started the AI race”