男女纠察队
Beijing launches flood response as heavy rain looms_我的网站

一 | 8 月 24 日消息,据外媒 Phys.org 今天(24 日)报道,卡迪夫大学和墨尔本大学的一项新研究发现,生成式人工智能(GenAI)目前还无法像人类教师一样可靠地评判学生的书面作业。研究人员使用 ChatGPT 的两个版本,对 50 篇生物科学专业本科生论文进行评分,以测试大模型评估学生作业的能力。

Photo: VCG
Beijing activated a Level II flood-control emergency response, the second-highest level, on Tuesday as authorities warned of heavy rain and heightened risks of urban waterlogging and landslides.
Most parts of the city are expected to receive more than 100 millimeters of rain within six hours and over 150 millimeters within 24 hours from Tuesday evening through Thursday, according to the Beijing meteorological service.
Beijing issued an orange alert for rainstorms at 10 a.m. Monday, while several districts, including Fangshan, Daxing and Tongzhou, issued the highest-level red alerts.
Authorities said residents in high-risk areas would be evacuated by Tuesday evening. Construction work will be suspended, and mountain campsites and scenic areas will be closed.
Starting Wednesday, employers are encouraged to adopt flexible working hours or staggered schedules. Schools may suspend classes and in-person training depending on conditions, while organized outdoor sports activities will be halted.
Beijing's drainage authorities have put 252 flood-control teams on standby. Preparations, including the installation of flood barriers, have also been stepped up at 368 subway stations, 33 hospital campuses and 17 elderly care facilities.
In Fangshan, where exceptionally heavy rain is forecast, authorities have begun evacuating residents from high-risk areas ahead of the downpours and deployed emergency rescue teams and supplies, with more than 150,000 items of emergency equipment and materials on standby.
Dolphin, the 13th typhoon of the year, made two landfalls in Zhejiang Province on Sunday. The National Meteorological Center on Tuesday maintained an orange alert for rainstorms, forecasting heavy rain across a broad swath of the country through Wednesday afternoon, including Zhejiang, Shanghai, Shandong, Beijing and Tianjin.
As of 7 a.m. Monday, about 167,900 people had been moved from vulnerable areas on land and at sea in Fujian Province. By late Sunday, Shanghai had evacuated around 215,600 people from areas considered at risk.
。ChatGPT 需要按照 7 项标准批改这些论文,研究人员还采用了 4 种不同的提示方式,随后将大模型与人工评分的平均分和分数波动情况进行比较。图源:Pexels卡迪夫大学生物科学学院威廉 · 凯博士指出:“研究结果显示,模型给出的分数波动很大,无法准确预测人工评分。生成式 AI 与人类批改同一批论文时存在明显差异。只看论文总分,两者给出的结果相对接近;一旦具体到每项评分标准和每一篇论文,大模型与人工评分之间的差距就相当明显。

二 | ”除一种情况外,AI 模型给出的平均分普遍高于人工评分。

三 | 平均分方面,AI 模型与人工评分的最大差距达到 16.1 分;具体到单篇论文,最大差距达到 40 分。

四 | 获悉,威廉 · 凯还发现,AI 往往会压低高分论文的成绩,又把低分作业的成绩抬高,最终使分数系统性地向中间集中。研究结果说明,至少现阶段,即使经过大量训练,AI 仍无法可靠地对主观性较强的书面作业给出与人类相当的分数。

五 | “我们测试的大模型目前并不适合取代教师,为学生预测作业成绩。”研究人员指出,高等教育领域确实很关注大模型能否利用模式识别能力,让学生作业评分更加客观,同时提高批改效率、减轻教职人员压力。但这项研究表明,目前并不适合这么做。

六 | 此外,未经学生明确同意便把作业提交给 AI 工具,本身还涉及伦理问题;大模型也不能、也不应该被依赖来给学生的长篇书面作业评分。

七 | “随着大模型继续发展,未来模仿人类判断的能力可能会提高。但从这项研究来看,要让 AI 给出的分数与人工评分真正保持一致,恐怕并不容易。
Current article:http://orgct.tourunbiaodubaicubangninan.shop/6cwey/gy4jb.html
Published on:11:15:32