[State of Research Funding] Beyond NSF, Slingshots, Open Frontiers — Andy Konwinski, Laude Institute
TL;DR
Andy Konwinski details the Laude Institute's dual-track model to revolutionize research funding by combining a nonprofit for open-source AI with a venture arm for PhD-led startups, aiming to replicate the 'Databricks Motion' where breakthrough research becomes billion-dollar companies through high-velocity, Silicon Valley-style capital deployment that complements underfunded NSF programs.
🏛️ The Laude Institute Architecture 3 insights
Dual-track funding model
The Institute operates a nonprofit for pre-incorporation open research and Laude Ventures for post-incorporation startups, unified by an obsession with 'shipping'—open source projects for the nonprofit and commercial products for the venture arm.
Researcher-led governance
Organization runs 'for researchers, by researchers' with 50+ top faculty (including Jeff Dean) and unicorn founders from Databricks and Perplexity serving as investment partners and advisors.
Right resource, right time
Structure targets the critical 'dotted line' of incorporation to provide resources exactly when researchers need them, bridging the gap between academic breakthrough and commercial application.
🚀 The Research-to-Unicorn Pipeline 3 insights
Multi-founder teams derisk startups
Large research teams with 7-8 co-founders (like Databricks, OpenAI, or VMware) reduce 'founder divorce risk' while proving ability to execute breakthroughs together before raising capital.
Replicating the Databricks Motion
Path follows NSF-funded research becoming open-source breakthroughs (like Apache Spark) that translate into hundred-billion-dollar companies, a model also seen in Perplexity, Anysphere, and Google.
New gold standard for VCs
Researcher-led startups have reached a tipping point where PhD founders are becoming the highest-leverage path from breakthrough to world-changing company, attracting top-tier venture interest.
🧠 Funding 'The Layer Above' Models 3 insights
Compound systems focus
Slingshots fund the 'layer above' base models—context management, agent tool use, memory curation, and prompt optimization—exemplified by DSPy/JEPA optimizers using evolutionary genetic techniques.
Post-training efficiency
Research targets continual learning methods that update model knowledge without full retraining (avoiding 10,000 GPU hours), analogous to human memory updating rather than brain rewiring.
Concrete funded projects
Specific initiatives include Terminal Bench (evaluation frameworks), Arena (comparison platforms), and research into compound AI systems that manage context and agent decisions.
💵 Complementing Traditional Funding 3 insights
NSF's critical funding gap
NSF's $1 billion annual computer science budget—facing potential cuts to half—is insufficient for frontier AI research, which requires $10-100 billion annually to compete globally.
Venture-style selection
Replaces traditional peer review with Silicon Valley 'picker' expertise to deploy capital faster, with stronger opinions narrowly focused on high-impact AI systems, cryptography, and architecture.
Geographic diversification
Expanding beyond Berkeley/Stanford to CMU, MIT, UIUC, University of Washington, and Canadian universities (Toronto, Waterloo, McGill) via PhD entrepreneurship clubs.
Bottom Line
Research funding must evolve beyond traditional grants to adopt high-velocity, venture-style selection of researcher-led teams working on open-source breakthroughs, creating a replicable pipeline from academic discovery to world-changing companies.
More from Latent Space
View all
🔬There Is No AlphaFold for Materials — AI for Materials Discovery with Heather Kulik
MIT professor Heather Kulik explains how AI discovered quantum phenomena to create 4x tougher polymers and why materials science lacks an 'AlphaFold' equivalent due to missing experimental datasets, emphasizing that domain expertise remains essential to validate AI predictions in chemistry.
Dreamer: the Agent OS for Everyone — David Singleton
David Singleton introduces Dreamer as an 'Agent OS' that combines a personal AI Sidekick with a marketplace of tools and agents, enabling both non-technical users and engineers to build, customize, and deploy AI applications through natural language while maintaining privacy through centralized, OS-level architecture.
Why Anthropic Thinks AI Should Have Its Own Computer — Felix Rieseberg of Claude Cowork/Code
Anthropic's Felix Rieseberg explains why AI agents need their own virtual computers to be effective, arguing that confining Claude to chat interfaces severely limits capability. He details how this philosophy shaped Claude Cowork and why product development is shifting from lengthy planning to rapidly building multiple prototypes simultaneously.
⚡️Monty: the ultrafast Python interpreter by Agents for Agents — Samuel Colvin, Pydantic
Samuel Colvin from Pydantic introduces Monty, a Rust-based Python interpreter designed specifically for AI agents that achieves sub-microsecond execution latency by running in-process, bridging the gap between rigid tool calling and heavy containerized sandboxes.