Honesty.Tools / Notes from the work
Personal accounts and practical guides, with sources and limits attached.
Audit notes
MIT CSAIL showed an AI editing a small model's weights and lifting its MATH-500 score from 7.9% to 48.1%. We checked the released records: the gains hold up, the edit rebuilds exactly, some failures remain, and telling models apart is not yet explaining them.
Read the articlePersonal account and practical guide
OpenAI now runs activation probes on every GPT-5.6 completion. What their post says they do, credit for publishing it, and the questions a fair outside check would ask: selection and seeds, drift, shortcuts, harmless look-alikes and numbers tied to runs.
Read the articlePersonal account
Nikita Pavlichenko tested LLM design changes since GPT-3 under one budget, stacked the winners, and listed his limits first. What a fair comparison looks like, and what it changes for our own work.
Read the articlePersonal account
HLE batch 2 passed, and its grouped test only just did. What we fixed in advance, why batch 1 was voided rather than fixed, and what the result can't tell you.
Read the articlePractical guide
How our receipt log works in plain words: signed checkpoints, an append-only log, Rekor anchoring and two outside witnesses. With commands to check it yourself in a few minutes.
Read the articlePersonal account and practical guide
An unexpected attachment led to a practical question: who can authorize an agent to act? A morning investigating prompt injection, permissions and exact-action approval.
Read the articleNotes from a re-check
We re-checked Goodfire's SDXL-Turbo feature 7446 claim, and it held up. Two of three follow-up features passed, we found two small loader bugs and offered a fix, and two locked studies have no results yet.
Read the article