Observed arrival · 2026-09-28
ENGRAFT: removable facts for a language model
ENGRAFT is an experimental method for writing facts into selected rows of a language model’s n-gram table, without changing the model’s weights.
○Open
⊠Login
$Paid
†Ads
✦Pretty
●Pro
◎Niche
◉Human
⚑Risk
ƒJS
- For
- Language-model researchers and inference-engine developers
- Worth noticing
- On one model, the site reports 707 exact answers out of 841 held-out test sentences, while noting the test uses corpus-like phrasing.
Field notes
The page separates exact-answer testing on corpus-like phrasings from free-form questions, reporting 707 of 841 for the former and 35 of 98 for the latter. It also measures changes on neutral text against the table’s own 4-bit storage noise. The simulator is illustrative rather than a live view of actual hashes, and the project identifies cross-subject composition, crowding, updates, and comparisons with other methods as unmeasured.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
○OpenPublic substance visible
✦PrettyNotable craft visible
●ProPolished or operationally mature
◎NicheUnusually specific use
One card from the complete issue