If the last decade of AI has taught us one lesson, it's that scaling compute builds better models.
【逻辑漏洞】文章开篇即确立了「算力决定论」的框架，但这是一个高度可争议的前提。DeepSeek-R1 用远低于对手的算力取得竞争性成果，恰恰说明算法效率可以部分替代算力——作者用这个反例贯穿全文，却又在框架层面偷偷把它收编为「几倍效率提升，不够弥补十倍差距」。这种循环论证让结论在逻辑上显得比实际上更无懈可击。
( ~5:00 ) Reading Aids should be used after initial interpretation. This is to prevent framing bias.
Bressan, P. (2021). Strangers look sicker (with implications in times of COVID-19). PsyArXiv. https://doi.org/10.31234/osf.io/x4unv
Taber, J. M., Thompson, C. A., Sidney, P. G., O’Brien, A., & Updegraff, J. (2021). Experimental Tests of How Hypothetical Monetary Lottery Incentives Influence Vaccine-Hesitant U.S. Adults’ Intentions to Vaccinate. PsyArXiv. https://doi.org/10.31234/osf.io/ux73h
Agarwal, A. (2021). The Accidental Checkmate: Understanding the Intent behind sharing Misinformation on Social Media. PsyArXiv. https://doi.org/10.31234/osf.io/kwu58
The framing effect, which is the bias the above examples exploit, is in my opinion the most dangerous bias in the world.
if packages and their elements are essential tools, then it makes a considerable difference that some are more readily available than others. Making sense of the world requires an effort, and those tools that are developed, spotlighted, and made readily accessible have a higher probability of being used
i.e. the most available and accessible frames are the most likely to influence public opinion