LEADERSHIPOngoing practice85% confidence

Research Leader as Portfolio Allocator

A research org wins by concentrating compute on a few big bets and disbanding the rest — not by letting a bottom-up credit market allocate it.

Problem it solves

Why a resource-rich research org with the best people and the founding technology can still lose the scaling race.

Best for

Research and engineering leaders allocating scarce compute and talent across competing internal bets.

Not ideal for

Genuinely exploratory phases where the winning direction is unknown and breadth is still buying information.

Overview

Why this framework exists

Luan frames the research-leader job as a portfolio allocator: coalesce smart people around a small number of good ideas, run them to the finish line, then nudge resources toward what is working and disband what is not. The named anti-pattern is Google's "Brain Credit Marketplace" — every researcher held compute credits, so a giant training run required convincing 19–20 colleagues to give up theirs. That bottom-up structure stopped Google reaching GPT-3 critical mass despite inventing the Transformer and having Noam Shazeer pushing trillion-parameter models. OpenAI won "simply because we took big swings and we focused."

Core principles

3 total
  1. Treat leadership as portfolio allocation: back a few ideas, run them to the finish, cut the rest.
  2. Bottom-up compute credit markets prevent the critical mass that scaling laws reward.
  3. Big swings plus focus beat a thousand under-resourced ideas.

Origin story

How this framework came to be

Luan led Google's LLM effort and co-led Google Brain for a year, then contrasted that bottom-up structure with the focus-and-concentrate model he pushed as a research leader at OpenAI.

Source

Traced to primary
Source · PODCAST
Why Google failed to make GPT-3 + why Multimodal Agents are the path to AGI — with David Luan of Adept
Latent Space (swyx & Alessio) · 2024
Open source →

Related frameworks

Browse all Leadership →