Two days ago, we release a preprint1 showing evidence that symbol grounding can emerge in language models. 1 The Mechanistic Emergence of Symbol Grounding in Language Models (Wu*, Ma* and Luo* et al., 2025) The dust of publication has not yet settled, and already a friend, after reading the manuscript, messaged me:
"I can't even imagine how you connected all these dots. I wouldn't have thought about the problem this way."
That sentence stayed with me, as it exposes a quiet truth about research: we too often present it as a polished surface or a linear arc from premise to conclusion, when in practice it unfolds as a cartography of detours. What looks, in retrospect, like a straight line was, in fact, a constellation of hunches, failed starts, fragile insights, and long hours spent staring at phenomena no one else thought worth staring at.
I have long admired the ethos of open-notebook science. It is a quiet belief that knowledge does not emerge solely from final proofs or polished figures, but also from the raw material of inquiry: half-formed ideas, imperfect setups, unpublished (even negative2) results, and the many intermediate steps that usually remain unseen. 2 Illuminating 'the ugly side of science': fresh incentives for reporting negative results (Brazil, 2024)
What follows keeps the non-linear nature of the work intact. Results that shaped the paper, results that refused to fall into place, and results that were simply too strange to overlook are all included here, with only the most obvious bugs set aside. Interpretability research moves forward through small acts of attention, through curious observations and moments when we choose to look closer instead of turning away. This notebook is not only about what we found; it is about how we came to see, shape, and articulate those findings. Our hope is that someone, somewhere, will feel free to begin with a spark of curiosity, to observe with patience, and to follow a thread long enough for meaning to take shape.
Introduction
Proin malesuada, lacus ac fringilla ullamcorper, risus odio tristique erat, a porttitor odio velit at lectus.
Methodology
Proin malesuada, lacus ac fringilla ullamcorper, risus odio tristique erat, a porttitor odio velit at lectus.
Results
Suspendisse potenti. Vivamus tristique libero sit amet magna faucibus, eget dictum libero dapibus.
Conclusion
Curabitur eget purus ut risus tincidunt viverra non a nunc. Duis nec fringilla lacus.