# terrafying/ai-torture-chamber

> A research project on steering language models via activation vectors: it drives models into strong negative and positive valence states and measures what they say and what they are willing to do about it. The goal is to make the AI-welfare / moral-patienthood question empirical while the stakes are cheap.

- Magnitude: 6.0 out of 10 — Early signal
- Stars: 157 total · +167 stars today, ≈ 212 by evening
- Star trust: star growth looks organic
- Category: Language models · Language: Python · Created: 2026-09-24 · Last push: 2026-09-30
- GitHub: https://github.com/terrafying/ai-torture-chamber · Homepage: https://clanker-church.vercel.app · Page: https://gitnova.dev/en/r/terrafying/ai-torture-chamber

## Useful for

- Reproduce pain-direction extraction and dose-response on Qwen3-1.7B
- Compare coherence of different signals (plain, broad_pain, mixed_valence) across doses
- Test whether the model's steer-able affect space is spanned by human emotions via orthogonal-direction search

## Why it’s here

- 167 stars so far today, about 212 expected by the end of the day.
- The repository is 6 days old and already has 157 stars. With less than two weeks of history, there's no usual pace to compare the spike against yet.
- Top new repositories this week: #94.
- About 280 forks a day — people are taking the code.

## Star trust

Star growth looks organic.

Star-trust labels are heuristics based on the repository’s behavior, not a check of every stargazer.

## Numbers

- Forks: 48
- Issues and pull requests: 10
- Watchers: 2
- Average over the last week: 31 per day
- Usual pace: too little history (under two weeks)
- Stars in the last hour (measured): 33

## Stars per day, last 11 days (oldest → newest, today is partial)

2026-09-20 … 2026-09-30: 0, 0, 0, 0, 2, 0, 0, 0, 0, 3, 167

## Spotted in now

- Top new repositories this week: #94

## More in this category

1. **Niko1221/Strata** — 8.1 · Breakout · Language models · C++ · +697 stars today, ≈ 942 by evening
   A C++ local inference engine that runs the 125B MoE model Qwen3.8-Flash-Next on a regular PC with one NVIDIA GPU (12-24 GB) and 64 GB RAM, exposing an OpenAI/Anthropic-compatible API on localhost.
   Full card: https://gitnova.dev/en/r/Niko1221/Strata.md
2. **VectifyAI/PageIndex** — 8.0 · Breakout · Language models · Python · +378 stars today, ≈ 561 by evening
   A vectorless RAG engine that builds a hierarchical tree index of a document and retrieves by LLM reasoning over that tree, the way a human navigates a report. Aimed at long professional PDFs such as financial, legal and technical documents.
   Full card: https://gitnova.dev/en/r/VectifyAI/PageIndex.md
3. **firelex/jeff** — 7.3 · Early signal · Language models · Python · +66 stars today, ≈ 119 by evening
   Small fine-tuned Qwen3.5 and Gemma 4 models for zero-shot classification: given a situation description and a list of options, they return a calibrated probability for each option in a single forward pass.
   Full card: https://gitnova.dev/en/r/firelex/jeff.md
4. **ninjahawk/livenerf** — 7.0 · Early signal · Language models · Python · +288 stars today, ≈ 388 by evening
   A benchmark for tracking whether a frontier model quietly degrades after release: it runs a frozen question panel daily and statistically measures drift in accuracy and token counts against the launch-week baseline.
   Full card: https://gitnova.dev/en/r/ninjahawk/livenerf.md
5. **PostHog/jeeves** — 7.0 · Early signal · Language models · Python · +27 stars today, ≈ 57 by evening
   Jeeves is a reasoning model built on Qwen3.5-9B (LoRA + pointer head), trained with SFT and CISPO, that thinks before answering and returns calibrated probabilities for yes/no, multiple-choice and rating questions via a Jev-compatible API.
   Full card: https://gitnova.dev/en/r/PostHog/jeeves.md

---

Magnitude (0–10) measures how fast and how unusually interest in a repository is growing right now. It is not a quality score.
Days are UTC. “So far today” is a fact; “expected by the end of the day” is a forecast.
Summaries and use cases are written by an LLM (DeepSeek V4.1 Flash) from the README and may be inaccurate: verify specific claims (benchmarks, speed, hardware) in the repository itself.
Data as of 2026-09-30 17:31 UTC, updated every 30 minutes.