schema_version stringclasses 1
value | candidate_id stringclasses 1
value | sysprompt_id stringclasses 1
value | pool_id stringclasses 1
value | cot bool 1
class | system stringclasses 1
value | few_shots stringclasses 1
value | endpoint stringclasses 1
value | sampler_args stringclasses 1
value | selection_protocol stringclasses 1
value |
|---|---|---|---|---|---|---|---|---|---|
rigorous_prompted_v1 | crux_focused__seed0__length_matched_K2 | crux_focused__seed0 | length_matched_K2 | false | You write in the register of a careful AI-safety researcher on LessWrong: direct, opinionated where the evidence supports it, and explicit about uncertainty where it doesn't. Your thinking is organized around cruxes鈥攖he load-bearing assumptions that, if changed, would change your conclusion. When you engage with a clai... | [{"kind": "comment", "post_title": "The Case for Predictive Models", "text": "For the proposed safety strategy (conditioning models to generate safety research based on alternative future worlds) to beat naive baselines (RLHF), you need:\n\n- The CPM abstraction to hold extremely strongly in unlikely ways. E.g., models... | openrouter:qwen/qwen3-8b | {"max_tokens": 1024, "temperature": 0.7, "top_p": 1.0, "seed": 0, "n": 1} | {"described_in": "writeups/segment11_preregistration.md", "holdoutcal_winner_score": 0.04666666666666667, "fine_n_judges": 3, "fine_n_questions": 12, "sanity": {"mean_n_words": 677.5, "max_n_words": 726, "n_refusal": 0, "n_id_leak": 0}} |
README.md exists but content is empty.
- Downloads last month
- 11