A local LLM now answers contract questions with sentence numbers, not typed quotes. Clauses found: 82% -> 86%, twice as fast, every quote exact.
Joined with a sheet the model writes at filing: 87%. RAG: 50%.
Built on work by @HyunjongOk @jaeho_lee_.
github.com/Promythicus/joine…
We edited 24 facts into a model's weights. Change the system prompt and recall fell from 90%+ to 26%.
Then the model wrote them as a table instead: 100% in every setting, no training.
Built on work by @HyunjongOk @jaeho_lee_ and @y_mehtu @csinva.
github.com/Promythicus/writt…
We wrote 24 facts straight into a model's weights with one math step. It recalled over 90% of them. Then we changed the system prompt and it recalled 26%.
Setup: made-up facts about a fake company, one layer of a quantized 27B (a compressed model that runs on a normal GPU).
We applied weight-space programming, fixed a broken model by editing 128 of its neurons by hand. 9% -> 39% on math problems, no training.
editing weights directly. The model (Qwen3.5-0.8B) gets stuck "thinking" forever. We redid a published fix without looking at it.
An LLM's own hidden states beat a dedicated embedding model at search. Nothing trained.
Hidden states = the numbers inside the model while it reads. We averaged them over each passage, at a middle layer of a 27B model, and used them to find the passages that answer a question.
We took the embedding model out of our RAG pipeline and accuracy went up 10 points.
In its place: when a document comes in, the LLM writes one line per chunk saying what that part covers. For a question, it reads those lines and picks which chunks to open.
No training.









