Which AI stories has SIREN scored in this window? · What has SIREN scored under Governance? · What has SIREN scored as Forums? ·
The Snitch Reflex: Can We Force Language Models to Be Compulsively Honest?
I fine-tuned a toy model to vomit back arbitrary tokens. I used them as Markers tied to specific subsets of its training data to connect training to generation. Training + testing cost $0.49 on an…
Source: What has SIREN scored via LessWrong? · original · first seen by this index at
Publisher: What has SIREN scored from lesswrong.com?. A publisher link means the original URL host cleared the outlet gate (≥5 scored stories in this window).
Topics: What has SIREN scored about AI safety?, What has SIREN scored about Reasoning models?, What has SIREN scored about Fine-tuning?, What has SIREN scored about Training data?, What has SIREN scored about Chain-of-thought?. A topic link means the title or summary matched a published needle with enough other stories to clear the tag gate.
What is the score on this story?
| Component | Points |
|---|---|
| recency | 19.5 |
| source weight | 12.5 |
| corroboration | 0.0 |
| entities | 0.0 |
| engagement | 0.0 |
| severity | 0.0 |
| coverage | 0.0 |
| Total | 32.0 |
Every term is published and every input is public, so this number can be recomputed. It ranks the item inside the 800-item window; it is not a judgement about importance in the world.
How many sources carried this story?Corroboration
Carried by one source. That is the normal case — most items are reported once — and it is stated rather than left blank so a single-source item cannot be mistaken for a corroborated one.
Which related stories are in this window?
- Notice of Revisions to Performance Area One of the LSC Performance Criteria 37.5
- Privacy Act of 1974; System of Records 37.5
- Was the HuggingFace hacking incident scheming? 53.4
- OpenAI Faces Lawsuits, Subpoenas and EU Scrutiny in a Single Week 36.5
- Disproving the SIA disproves the Fine-Tuning Argument and Boltzmann-brain-based arguments 33.7
What questions do people actually ask?
What is the score on this story?
Scored 32.0 of 100. Every term is published and every input is public, so this number can be recomputed. It ranks the item inside the 800-item window — not a judgement about importance in the world and not an item doom score.
How many sources carried this story?
Carried by one source. That is the normal case — most items are reported once — and it is stated rather than left blank so a single-source item cannot be mistaken for a corroborated one.
Which related stories are in this window?
5 related stories in the same window, each with its own permanent scored page.
Does SIREN publish an item doom score?
No. Activity tempo, not probability of harm. A permanent story page carries a tempo score decomposition and corroboration set, not a probability of harm.