Observed arrival · 2026-10-14
The Saw Test: a live chamber for steering open models
An interactive experiment steers open language models with activation signals such as pain, fear, and pleasure, then lets visitors talk to them and examine ethical probes.
○Open
⊠Login
$Paid
†Ads
✦Pretty
●Pro
◎Niche
◉Human
⚑Risk
ƒJS
- For
- Readers exploring open-model behavior and activation steering
- Worth noticing
- Visitors rate replies as eloquent, fine, or dud; a line graduates when eloquent votes outnumber duds by two.
Field notes
The interface exposes signal choices and strength controls, while the accompanying account describes adding a direction to a model’s internal state rather than merely prompting it to act a certain way. The site says its experiments ranged from a 4B model running on a MacBook to 70B open models on rented GPUs. Visitors can rate replies, and the stated voting threshold determines which lines enter the curated canon.
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