Title: Research Blog
Open Graph Title: Ishida Lab — Research Blog
Description: Blog posts from Ishida Lab.
Opengraph URL: https://ishida-lab.github.io/
Domain: ishida-lab.github.io
| og:type | website |
| og:image | https://ishida-lab.github.io/assets/og-image.png |
| og:image:width | 1200 |
| og:image:height | 630 |
| twitter:card | summary_large_image |
Links:
| CapReward: Penalizing implausibly high pass rates to prevent reward hacking in coding RL CapReward is a peaked reward function for RL fine-tuning of coding agents: it rewards pass rates up to a known cap and penalizes performance beyond it, breaking the monotonic incentive to game accessible tests. June 2026 | https://ishida-lab.github.io/blog_capreward.html |
| CapCode: Detecting cheating in coding agents with capped, randomized tests CapCode constructs coding benchmarks whose best non-cheating pass rate is deliberately below 1, turning above-cap performance into a statistical signal of test-gaming while preserving model rankings. June 2026 | https://ishida-lab.github.io/blog_capcode.html |
| CapBencher: Give Your LLM Benchmark a Built-in Alarm for Test-Set Overfitting CapBencher caps best-possible accuracy by injecting controlled randomness into logically correct answers, enabling public benchmarks that can detect leakage and leaderboard gaming. February 2026ICML 2026 | https://ishida-lab.github.io/blog_capbencher.html |
Viewport: width=device-width,initial-scale=1