90.6% (1,587) have proved to be valid true positives, and 62.4% (1,094) were confirmed as either high- or critical-severity
这两个百分比数据点(90.6%验证率，62.4%确认高危率)对于评估AI模型在安全漏洞检测中的可靠性至关重要。90.6%的验证率表明AI模型的误报率相对较低，这在AI安全领域是相当出色的表现。然而，62.4%的确认高危率意味着近40%的AI评估高危漏洞实际严重程度较低，这反映了AI在严重性评估上仍有改进空间。
The NLA consists of the AV and AR, which, together, form a round trip: original activation → text explanation → reconstructed activation. We score the NLA on how similar the reconstructed activation is to the original.
NLA通过激活解释器(AV)和激活重构器(AR)形成闭环，通过重构质量评估解释准确性，这种创新方法为AI内部表示的可解释性提供了新范式。
We do not publish AI-generated images, audio, or video as authentic documentation of real events.
这条规定指出Ars Technica不会将人工智能生成的图像、音频或视频作为真实事件的证明，体现了对真实性的坚持。
Using these ability scores, the method predicts performance on new tasks with ~88% accuracy, including for models such as GPT-4o and Llama-3.1.
令人惊讶的是：ADeLe方法能够以约88%的准确度预测AI模型在新任务上的表现，这包括像GPT-4o和Llama-3.1这样先进的大模型。这种预测能力远超传统评估方法，为AI性能评估提供了革命性的突破，使研究人员能够更可靠地预见模型在未见过的任务上的表现。
We've seen customers go from 10-20% field accuracy with a frontier model to 99-100% just by switching to using Reducto's Deep Extract.
大多数人认为从前沿模型到接近完美的准确率需要根本性的技术突破或大量数据训练。但作者声称仅通过切换到Deep Extract方法就能将准确率从10-20%提升到99-100%，这种巨大性能提升的幅度与行业通常预期的改进曲线相悖，暗示现有方法可能存在根本性缺陷。
ReconfigBehSci. (2020, November 9). Session 2: The policy interface followed with a really helpful presentation by Lindsey Pike, from Bristol, and then panel discussion with Mirjam Jenny (Robert Koch Insitute), Paulina Lang (UK Cabinet Office), Rachel McCloy (Reading Uni.), and Rene van Bavel (European Commission) [Tweet]. @SciBeh. https://twitter.com/SciBeh/status/1325795286065815552
Yang, Scott Cheng-Hsin, Chirag Rank, Jake Alden Whritner, Olfa Nasraoui, and Patrick Shafto. ‘Unifying Recommendation and Active Learning for Information Filtering and Recommender Systems’. Preprint. PsyArXiv, 25 August 2020. https://doi.org/10.31234/osf.io/jqa83.