LLM capabilities can transfer through unrelated text
A new study demonstrates that Large Language Model capabilities can transfer to unrelated tasks through a method called Active Taskless Distillation. The research shows that training on single-word responses to task-unrelated prompts significantly improves performance on coding benchmarks compared to exact controls. The authors have released code and data to allow independent verification of these results using public models.
- ▪The study found that the signal group achieved a 51.22% pass@1 rate on HumanEval+, which is 5.34 percentage points higher than the 45.88% rate of the exact control group.
- ▪The experiment utilized the Qwen2.5-1.5B-Instruct model and required only public data, eliminating the need for a private teacher model during student training.
- ▪Researchers provided a reproducible pipeline that verifies file hashes and recomputes statistical results from frozen training arms to ensure transparency.
- ▪The training process involved 157 optimizer steps with LoRA rank 16 and a learning rate of 5e-5, using single-token full-vocabulary cross-entropy loss.
Hacker News (AI / LLM) files mainly under ai. We currently carry 6,661 of its stories.
Story provenance
Source · retrieval · rights · ranking — open for full record
inspect →
Story provenance
Attribution is not the same as permission. This drawer separates discovery metadata, excerpts, WeSearch-generated summaries, reuse status, and whether the publisher receives the visit. Nothing here claims a legal grant the publisher has not made.
Record
| Original publisher | GitHub |
| Canonical URL | https://github.com/myboker/ATD |
| Publication time | Mon, 28 Sep 2026 06:11:43 +0000 |
| Retrieval time | 2026-09-28T06:26:07.564Z |
| Last seen | 2026-09-28T06:26:07.564Z |
| Headline source | Publisher (no WeSearch rewrite) |
| Excerpt source | publisher body |
| Excerpt method | First ~120 words (~800 chars) of extracted publisher body, fair-use limited. |
| Summary | WeSearch · cerebras-chat (WeSearch summarizer) |
| Summary source text | contentText |
| Citation coverage | Summary is a WeSearch-generated derivative; primary citation is the original publisher URL. |
| Cluster | SPojEtOiIGi0 · 1 stories |
| Cluster logic | Grouped by semantic title/content similarity across sources within a rolling window. Same-publisher template collisions are excluded from coverage comparison. |
| Ranking reason | Story pages are not engagement-ranked. Hub feeds use recency, with optional source-diversified chronological ordering (cap consecutive stories per source). No personalized ranking. |
| Publisher visit | Yes — open original |
| Substitutes article? | No — link-out required for full text |
Rights status (four layers)
WeSearch handling by dimension
| Indexing | May the item be indexed (stored, ranked, made findable)? | Allowed |
| Snippet | May a short excerpt of the publisher's text be shown? | Allowed |
| AI summary | May WeSearch generate its own short summary of the article? | Limited |
| Retrieval / RAG | May the content be exposed for third-party retrieval-augmented generation? | Not asserted |
| Model training | May the content be used to train AI models? | Not asserted |
| Commercial reuse | May the content be reused commercially? | Not permitted |
Basis: Derived from the published RSS/Atom feed. Contact: [email protected]. Reviewed: 2026-07-24.
Opening excerpt (first ~120 words) tap to expand
Active Taskless Distillation (ATD) Paper: arXiv:2609.29233 Reference code for the HumanEval+ signal vs. exact nuisance-matched control experiment in Post-Training Leaves Behavioral Shadows on Unrelated Decisions. Students learn from single-word responses to task-unrelated prompts. This release contains the two frozen training arms, their carrier metadata, and the four paired runs' full EvalPlus records. Student training requires only the public Qwen ancestor; no private teacher is needed.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at GitHub.