Metascience
Science is stuck
There's a lot to critique about the way we do science today. A few of the things keeping me up:
Research productivity is declining while paper counts explode — we're producing more noise than signal, and LLM-generated papers will only sharpen that curve. We measure citations as impact knowing full well it's a poor proxy. Much of the published record is not reproducible, yet we keep funding it and building on top of it. The pressures of the paper paradigm reward findings dressed up as more certain than rigorous assessment would warrant — and leaning on LLMs, those famously over-confident creatures, for literature review only makes it worse.
We're producing more noise than signal.
A new data model
None of this changes until we change the way we represent and contribute knowledge. Oshima is building a better unit of knowledge for science. We're exploring new ways to represent the literature that map closer to the scientific method and let us measure and reward research more truthfully. Inspired by the empirical method itself, we think in terms of claims and evidence. Picture a vast, collaboratively built knowledge landscape: researchers contribute new claims, they and others attach evidence, and everyone is recognized for the research they actually did.
As long as we stick to papers, we keep lumping together very distinct parts of the scientific method — and stay unable to recognize and reward them on their own.
Product
Drill down to the evidence
Modular breaks papers into claims, experiments, and evidence — and shows how they connect across the literature. It's the first piece of Oshima you can use today.
Try Modular
Writing
Notes on why the paper is broken and what could replace it. carlaostmann.substack.com
- Jun 2026 Alien proteins, or: what if the cure is something nature never made? Embedding all of biology into one space — and generating proteins that don't exist yet. New forms of knowledge representation for automated discovery. Read on Substack
- Apr 2026 Papers block scientific progress Why the paper — and the citation count we grade it by — became the bottleneck stopping science from discovering faster. Read on Substack