Signal Net

  • The Frozen AI Problem: How Researchers Are Teaching LLMs to Keep Learning

    The AI tools you use every day — Claude, ChatGPT, Gemini — are, in a meaningful sense, frozen in time. They learned everything they know during a multi-month training run, and then the learning stopped. New research published in 2025 and 2026 is attacking this problem from five different angles simultaneously. Here’s what’s actually working,…

  • The papers being cited fastest in 2025 are not discoveries

    SciPy 1.0 was published in February 2020. In the twelve months ending October 2025 it was cited 1,731 times. That is a five-year-old methods paper picking up roughly five new citations per day, with the rate still climbing. Its acceleration over the prior year is +94 per month, the third highest in the entire top-300…

  • The research papers on fire right now, graded month by month

    Most “trending papers” lists are annual cuts: how many times a paper got cited this year versus last. That misses the shape of the curve. A paper that quietly clocked 200 citations evenly across twelve months looks identical to one that went from 5 per month to 40 per month. The second is actually taking…

  • What one ESC guideline is doing to the cardiomyopathy drug pipeline

    In 2023 the European Society of Cardiology published its first comprehensive guideline on cardiomyopathies. OpenAlex has already logged more than 2,100 citations to that single document, a pace that puts it among the fastest-accelerating clinical papers of 2024. The guideline itself is not the story. The story is what it quietly codified: a drug class…

  • Plotting Thiel: Nineteen Long-Baseline Charts Test the Stagnation Thesis

    Steven Pinker’s Enlightenment Now proves that, on a 200-year baseline, almost everything is up and to the right. Peter Thiel’s The End of the Future argues that, since 1973, almost nothing in the physical world is. Both can be true. The trick is to plot each metric on its own longest available baseline and see…

  • In the Desert, Not the Enchanted Forest: Re-Reading Thiel and Kurzweil From Inside a Nine-Million-Patent Corpus

    Peter Thiel said in 2011 that the future had stalled. Ray Kurzweil said in 2005 that the future was on schedule. Reading both texts now, with a 9.3-million-patent corpus running underneath them, the answer is that they were arguing about different decades — and that the inflection between those decades happened, with disconcerting precision, the…

  • The Problem Graph: From Patent Tinkering to an Idea

    What started as a search for hidden innovation in medical devices turned into something bigger — a vision for how AI-driven R&D will need to coordinate across fields, and the infrastructure that doesn’t exist yet.