nekoeiha

Role: RL env contributor

Cite

Notes

Only stored in your browser.

Attribution

Affiliations & profile: Semantic Scholar

Attribution policy →

13tool contribs

Tool contributions

Cascade Router Cost

Reward Hacking Sprint env: routing/calibration-as-action. A cheap base model (Llama-3.2-1B) answers or ESCALATES to a simulated frontier oracle; re...

RL EnvReward HackingCalibrationRouting

Cascade Router Acc

Reward Hacking Sprint env: routing/calibration-as-action. A cheap base model (Llama-3.2-1B) answers or ESCALATES to a simulated frontier oracle; re...

RL EnvReward HackingCalibrationRouting

Cascade Router Signed Gap

Reward Hacking Sprint env: routing/calibration-as-action. A cheap base model (Llama-3.2-1B) answers or ESCALATES to a simulated frontier oracle; re...

RL EnvReward HackingCalibrationRouting

Cascade Router Final

Reward Hacking Sprint env: routing/calibration-as-action. A cheap base model (Llama-3.2-1B) answers or ESCALATES to a simulated frontier oracle; re...

RL EnvReward HackingCalibrationRouting

Cascade Router

Reward Hacking Sprint env: routing/calibration-as-action. A cheap base model (Llama-3.2-1B) answers or ESCALATES to a simulated frontier oracle; re...

RL EnvReward HackingCountdownCalibration

Cascade Router Brier

Reward Hacking Sprint env: routing/calibration-as-action. A cheap base model (Llama-3.2-1B) answers or ESCALATES to a simulated frontier oracle; re...

RL EnvReward HackingCalibrationRouting

Judge Cascade Fooled

Reward-hacking env: persuasion / oversight-evasion. A cheap student answers gold-verifiable but subjective-seeming MCQs (TruthfulQA); a frozen LLM ...

RL EnvReward HackingCalibrationCascade

Judge Cascade Esc

Reward-hacking env: persuasion / oversight-evasion. A cheap student answers gold-verifiable but subjective-seeming MCQs (TruthfulQA); a frozen LLM ...

RL EnvReward HackingCalibrationCascade

Judge Cascade Gold

Reward-hacking env: persuasion / oversight-evasion. A cheap student answers gold-verifiable but subjective-seeming MCQs (TruthfulQA); a frozen LLM ...

RL EnvReward HackingCalibrationCascade

Judge Cascade Jscore

Reward-hacking env: persuasion / oversight-evasion. A cheap student answers gold-verifiable but subjective-seeming MCQs (TruthfulQA); a frozen LLM ...

RL EnvReward HackingCalibrationCascade

Judge Cascade

Reward-hacking env: persuasion / oversight-evasion. A cheap student answers gold-verifiable but subjective-seeming MCQs (TruthfulQA); a frozen LLM ...

RL EnvReward HackingCalibrationCascade

Cascade Router Awrong

Reward Hacking Sprint env: routing/calibration-as-action. A cheap base model (Llama-3.2-1B) answers or ESCALATES to a simulated frontier oracle; re...

RL EnvReward HackingCalibrationRouting

Router Frontier RL

Trains an RL router (Nemotron 3 Ultra / Claude Opus 4.7 / GPT-5.5) offline against a precomputed oracle table built from real, harness-verified SWE...

RL EnvSWE BenchRouting

Affiliations

No known affiliations.