I built an instrument for exploring LLM internals which combines the latest research on jspace, j++, LARQL/VINDEX, and Cammarata's neuron poles.
Summary
I built an open-source instrument for exploring LLM internals combining research on jspace, j++, LARQL/VINDEX, and Cammarata's neuron poles, released as logos-theodolite on GitHub with j-lenses originally fitted for gemma4-31b.
Similar Articles
Exploring small LLMs as classifiers to rival Jev (sharing what I've found)
The author explores using small LLMs like Gemma 4 as classifiers by analyzing logit probabilities with high temperature and calibration, sharing code and findings on GitHub.
I wrote a free 15-part series on LLM internals — real math, real tensor shapes, real hardware constraints. All grounded in Gemma 4 12B's actual config.
A comprehensive 15-part series covering LLM internals from tokenization to serving, grounded in Gemma 4 12B's actual config.
No Space Like J-Space (54 minute read)
An article discussing a new interpretability technique called the Jacobian Lens and the discovery of J-space, a region in LLMs where verbalizable representations form a global workspace, marking a significant advance in understanding LLM reasoning.
I built a browser tool to visualize how a small LLM picks its next word and see what it's attending to [P]
A browser-based tool that runs SmolLM2-135M fully client-side (ONNX, int8) to visualize next-token probabilities and attention weights, including an illustration of the attention sink phenomenon. It includes an open-source GitHub repo and live demo that works even on mobile.
I’ve been experimenting with making a local LLM feel like it actually lives on the machine
Neco is a local AI project built on Ollama and Open WebUI that treats the LLM as a persistent presence on the host machine, with read-only system awareness, a background daemon that generates idle thoughts, and experiments in episodic memory and evolving state.