Automating Literature Research
FutureHouse
Winter 2024
FutureHouse Structure
- Non-profit
- Funded primarily by Eric Schmidt
- Based in San Francisco
- 20 employees
Science is changing independent of AI
Arxiv.org,10.6084/m9.figshare.17064419.v3
Intellectual bottlenecks are growing
📝 Increasing paper count ($\approx$5M per year)
🧬 Larger data sets from cheaper experiments (genome at
$200 per person, $1 / GB of sequencing)
🔍Increasingly less disruptive papers (96% decline in biology)
Mission
Automate Scientific Discovery
PaperQA: an agent for literature research
Language agents achieve superhuman synthesis of scientific knowledge
Michael D. Skarlinski, Sam Cox, Jon M. Laurent, James D. Braza, Michaela Hinks, Michael J. Hammerling, Manvitha Ponnapati, Samuel G. Rodriques, Andrew D. White arXiv:2409.13740, 2024
Better at answering questions than PhD biology experts
Better than competing models
Better than human written Wikipedia articles
Can detect if a claim is contradicted anywhere in literature
PaperQA is not ChatGPT
PaperQA is AI built on top of literature, rather than replacing literature.
We do not train LLM models on article text!
Applications
WikiCrow
- Wikipedia articles for all 19,255 protein-coding genes
- Succeeded on 17,269
- Wikipedia had 3,639, so gain of 13,630
- 48 Hours
Partnerships with FH
Interested in:
- Access to corpus to provide links to your content
- PaperQA for your corpus or topics
- Exploring research questions about interacting with literature