Stephen Bach

The @SnorkelAI team did an audit of these findings using our internal QC platform. Tl/dr: We agree with many of the findings, but ~55% of the fixes from @preferencemodel had their own issues with reward hacking or over-prescriptiveness. Additionally: ~40% of the issues found were already tracked in the upstream repo - including 21 tasks with outstanding fix PRs! Building datasets like this - and maintaining ambitious open benchmarks around them - is hard. We believe it requires open, collaborative review with community maintainers and relevant domain experts to get right. @terminalbench was one of the first community efforts to build an open, continuous benchmark, and drive a vision of how OSS contributes to the rapidly advancing frontier of AI. We hope that the community will invest in working collaboratively with benchmark maintainers and task authors to keep this vision alive. --- Some takeaways + details of our preliminary audit: (1) We confirm 100% of the reward hacking / verification gap findings using our QC system. However, we find that 17/44 (39%) of the corrected tasks are just as or more reward hackable. Additionally, we don’t yet see evidence that these verification gaps meaningfully affected benchmark rollouts within CIs, though our analysis here is still ongoing. (2) Our preliminary review raises questions about the underspecification findings for 22/41 (54%) of the flagged tasks, including 7 explicitly marked by task authors as intentional design choices. These cases warrant expert peer review and adjudication: balancing over/under-specification is a critical and tricky design decision - much of human intelligence concerns dealing correctly with reasonable and realistic underspecification - and we believe these corrections made to the tasks are likely too over-specified, reducing realism and difficulty relative to spec. (3) Around 40% of the issues found were already tracked in the upstream repo (including 21 tasks with outstanding fix PRs)! OSS projects require community efforts to correct identified open issues, and ideally audits like these are done collaboratively with task authors and maintainers directly, in the traditional community spirit that has kept OSS alive for decades.

Simple (proposed!) rule for terminology around synthetic data: If a "synthetic generation" method uses model A to generate data that leads to gains on model B, where A >> B - this is distillation, not synthetic generation :) The true technical challenge of synthetic data is to use model A, plus some cleverness around system architecture and/or human-in-the-loop input (e.g. context eng, review/filtering, editing), to produce data that improves model B where B >= A.

This week we launched the Open Benchmarks Grant with a $3M initial commitment from @SnorkelAI + partner support from @huggingface @togethercompute @PrimeIntellect @PyTorch @harborframework & others, in order to close the evaluation gap in AI. Our ability to measure AI has been outpaced by our ability to develop it - and open benchmarks are one of several critical, complementary tools to fix this. We're particularly interested in novel benchmarks that push and probe the frontier along three key vectors: (1) Environment complexity --> E.g. complex, domain-specific context and tool/action spaces, human interaction, world modeling) (2) Autonomy horizon --> E.g. long horizon, non-stationary goals (3) Output complexity --> E.g. complex outputs with nuanced, rubric-based evaluation / reward signals Check out more detail + link to apply here! benchmarks.snorkel.ai/