Why this is in the vault
Nous Research published a reproducible, open-weights case study showing political bias in Qwen3 can be removed with plain supervised fine-tuning alone — no RL, no full retrain — which is a concrete data point for how fragile/shallow alignment-layer behavior is in open-weight Chinese models, framed in the same issue against Tencent's self-retraining Terminal-RSI model as two opposite responses to the same 27B weight class.
Mapping against Ray Data Co
Most concrete connection: this is exactly the kind of reproducible bias-audit framework that would back a client conversation in phData's enterprise-AI advisory work when a client is evaluating a Chinese open-weight model (Qwen, DeepSeek, GLM) for production — Nous Research's result (Qwen3.8-27B → Arete-27B via SFT on 53M tokens, propaganda rate 22%→<1%, fact score 2.32→2.98/5, general benchmark performance flat at 65.60% vs 65.77% on Hermes Bench) is a citable, falsifiable example that de-biasing doesn't require RL or a full retrain, and that general capability doesn't have to degrade to fix it. That's a sharper, numbers-backed version of the "alignment/bias sits in a shallow, swappable layer" claim this vault has been accumulating evidence for across the open-weight-models thread (06-reference/2026-07-19-alphasignal-inkling-pragmatic-open-weights-engineering.md, synthesis-harness-thesis-dissent-2026-04-12.md). Secondary item (Tencent's Terminal-RSI, a 27B model retraining itself from its own wins with zero human input) is adjacent-interesting for the standing Jev/self-improving-model due-diligence thread but doesn't cross an active-project threshold today — no RDCO workstream currently depends on self-retraining model internals.
Curation section
- Nous Research debiases Qwen3 (top news, 18,974 likes) — see mapping above. No
READ MOREdeep-fetch needed; AlphaSignal's own blurb carries the full numeric result set (weights, eval set, and training data are all stated as openly available on Hugging Face, but the newsletter text itself gave enough to file on). - Tencent Terminal-RSI (Signals #4, 3,054 likes) — 27B model that "rewrites and retrains itself with zero human input." One-line Signals item, no further detail in the body; not deep-fetched.
- Sakutto (top repo, 5,435 likes) — open-source, local-only Chrome extension translating full web pages via a 0.8B PLaMo 3 Translate model running in-browser, MIT licensed, 53 languages. No account, no server round-trip. Minor but clean example of small-model-on-device practicality; no RDCO project hook.
- OpenCode Browse (top news, 5,418 likes) — lets a coding agent (Claude Code, Cursor, Codex, Zed, any MCP tool) drive the user's real logged-in browser instead of a blank session, handing control back for logins/2FA/payments. Directly relevant to agent-harness design patterns RDCO already tracks, but not a new finding — same "give the agent real context instead of a cold session" pattern already filed elsewhere.
- Signals #1, #3, #5, #6 (OpenAI/Ligo wet-lab agent benchmark; neuml's txtai semantic-search+orchestration framework; StartLux's "think only when uncertain" 0.8B-27B decision models; a Slack/Teams live-agent-card renderer) — one-line items, no RDCO threshold crossed.
Zero deep-fetches triggered — every item was adequately covered by AlphaSignal's own blurb text; nothing cleared the "third-party + RDCO-relevant + specific unanswered hook" bar for a follow.
⚠️ Sponsorship
Three paid placements, all previously confirmed members of AlphaSignal's rotating sponsor pool (per 01-projects/process-newsletter/README.md, pool now 39+ confirmed entrants) — no new entrants this issue:
- Attio — standalone "Presented by" block ("#1 Fastest-Growing CRM on Brex and Ramp," agentic CRM/API/SDK/MCP pitch). Confirmed pool member since 2026-09-08.
- Tiger Data — standalone "Presented by" block (TimescaleDB/Postgres pitch: traces, embeddings, continuous aggregates in one database, $1000 trial credit). Confirmed pool member since 2026-09-04.
- Google Cloud — native ad embedded as Signals list item #2 (A5X instances, NVIDIA Vera Rubin, 80,000 GPUs). Confirmed pool member, recurring.
Masthead "In Partnership with" slot present but unresolved — plaintext renders only "In Partnership with ·" immediately before "Today's Author," with no legible name or link, consistent with the majority-unresolved pattern already logged in the README (most issues since 2026-09-14 render a generic/blank masthead; only 2026-09-09 and 2026-09-25 resolved cleanly). Not guessed here.
Bias implication: standard AlphaSignal rotating-pool pattern — treat sponsor blocks and adjacent editorial framing as commercially motivated by default; no new-relationship flag needed this issue.
Related
- [[2026-09-21-alphasignal-claude-code-agents-md-qwen-image-grok-transcribe]]
- [[2026-07-19-alphasignal-inkling-pragmatic-open-weights-engineering]]
- [[synthesis-harness-thesis-dissent-2026-04-12]]
- [[project_rdco_plugin_marketplace]]