A group of master’s students at the Descartes Center launched the Journal of Trial and Error to normalize publishing negative and methodological findings and to bridge historians and philosophers of science with practicing scientists. Motivated by Open Science debates on replication failures and positive publication bias, they argue that science is fundamentally a trial-and-error process with two important kinds of errors: methodological errors that reveal practical limits and solutions across communities, and conceptual errors where hypotheses fail and spur theoretical revision. They diagnose three core problems - an idealized public image of breakthrough science, a gap between what is researched and what is published, and a replication crisis - and locate these within wider concerns about inaccessible publishing and disciplinary fragmentation.
The manifesto proposes a journal that publishes rigorous accounts of failed experiments and methodological pitfalls alongside reflective commentary from historians, philosophers, sociologists, and anthropologists. It emphasizes constructive reflection rather than sloppy research, aims to reduce publication bias and help reframe replication debates, and seeks to foster dialogue across specialties. The project has gained institutional traction - CrossRef membership, conference invitations, publisher interest, and submitted manuscripts - and calls for further participation from the scholarly community to make reflective practices customary.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.
Asaf Karagila criticizes OpenAI's preprint on the Partition Principle, calling it muddled, poorly structured, and inadequately referenced. He emphasizes the need for careful evaluation of AI's role in mathematical research and the importance of rigorous standards. (karagila.org)
OpenAI reportedly discovered a proof for the famous unique games conjecture, a major problem in computational complexity. Researchers raced to publish their own results to stay ahead of the machines' advancements. (quantamagazine.org)
A $1.8 billion global initiative is expanding efforts to create AI-ready biological data for disease research. Major partners like Biohub, NIH, DOE, Google DeepMind, and Meta aim to develop predictive models to accelerate understanding and treatment of diseases. (biohub.org)
Research suggests that ADHD may be linked to disruptions in circadian rhythms, indicating a potential role for chronotherapy in treatment. This approach could improve symptom management by aligning therapy with the body's biological clock. (frontiersin.org)
Model collapse, where AI models lose rare and unconventional data, is not new and affects many areas beyond AI. It results in homogenized outputs that lack the diversity and mistakes crucial for progress and discovery. (rulr.dev)
Scientists using ESA's Mars Express have discovered that Mars's Arsia Mons Elongated Cloud forms through a rare process called homogeneous nucleation, which does not require dust or particles. This exotic physics challenges previous understanding of Martian cloud formation and has only now been modeled successfully with advanced simulations. (esa.int)
Today's best Hacker News stories, summarized and screenshotted, one email a day.