Academia has a new preprints problem
Researchers are pulling the LLM slot-machine lever for novel physics, then wrapping the output in LaTeX. What that looks like from inside a preprint platform.
Topic
AI
What artificial intelligence needs from research infrastructure before it can do science, and what changes once it can.
Have we already hit the peer review breaking point?
Notes from the Royal Society's Future of Scientific Publishing: what does review look like when every paper is AI-generated or co-authored?
AI Needs New Facts – The Value of Novel Scientific Research
Hearing Demis Hassabis at SXSW London: if models train on what already exists, genuinely novel experimental results become the scarce input.
The Perpetual Research Cycle: AI's Journey Through Data, Papers, and Knowledge
Testing ChatGPT, Gemini and Claude at turning datasets into papers, and what a genuinely self-feeding research cycle would require.
The Role of Foundations and Technology Companies in Fuelling Optimism in Academic Research
As North American grant funding wobbles, Evo-2 and Google's AI co-scientist show where research optimism is now coming from instead.
Types of AI Agents and Their Applications in Academic Research
Model-based, goal-based, learning and utility agents: a practical taxonomy of AI agents and where each one fits in academic research.
Making poorly described data FAIR-er using GenAI
A small experiment: can ChatGPT-4 improve the metadata, and therefore the FAIR-ness, of badly described datasets already published openly?
Who benefits when, from FAIR data? Part 2 – Machines
AI and machine learning can be used to create more detailed, FAIR-er datasets to be consumed by the machines.
All of the knowledge for all of the machines
Two slow revolutions in publishing, blurred peer review and universal Open Access, decide what machines can legally read and actually trust.