查理和巧克力工厂

A |      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。    

    This photo shows a city view of Kampala, Uganda. Photo: XinhuaTravelers from the Taiwan region are being cautioned against traveling to Uganda after reports that several Taiwan residents had their travel documents issued by the Taiwan authorities seized or were denied entry in early August, amid a series of moves recently in Africa related to the one-China principle. A mainland expert noted that the more "Taiwan independence" forces seek international space, the narrower their room for maneuver will become.
According to Taiwan regional media outlet United Daily News (UDN) on Saturday, a Taiwan resident holding a valid Ugandan work permit had his so-called "passport" confiscated by Ugandan immigration officials on August 3 after arriving at Entebbe International Airport (EBB) on a domestic flight. The officials cited a "new policy" but have yet to return the "passport." 
In a separate but similar incident, another Taiwan regional resident was also denied entry by Ugandan immigration officials at EBB and deported on the same flight, according to UDN.
The two incidents occurred at the same airport within a short period of time, with the authorities' actions escalating from the confiscation of travel documents to outright denial of entry and deportation, a travel agency was quoted by the report as saying.
The developments suggest that the measures being taken by Ugandan authorities may go beyond isolated incidents and could represent a systematic pattern of enforcement targeting people from the region, with signs that the situation is "continuing to escalate," per the report by UDN.
However, so-called Taiwan regional foreign affairs authorities stated that the relevant cases may be connected with the measure rolled out by the Taiwan region to temporarily suspend visa issuance to Ugandan nationals and bar their entry amid the worsening Ebola epidemic, according to Taiwan's local media CNA, adding it will keep negotiating with Uganda and offer support to impacted people.
Prior to this incident, several African countries had already adjusted their wording and entry policies concerning Taiwan region. According to a previous UDN report in July, on the registration page for South Africa's Electronic Travel Authorization (ETA) system, Taiwan is listed as "Taiwan, China" in the dropdown menu for place of issue of travel documents. Similarly, the international dialing code +886 is marked "Taiwan, China" under the phone number section.
That same month, the Moroccan Embassy in Japan suspended its Visa Confirmation Letter (VCL) review process for travelers from Taiwan region starting July 23, while the regional authorities stopped accepting visa applications from Moroccan nationals in retaliation, according to CNA.
But some netizens from the island commented that they had "gotten used to this situation". One remarked, "It was Papua New Guinea a few days ago, and now Morocco again. Taiwan's foreign affairs department sure has an easier job these days."
After the government of Papua New Guinea announced the closure of the "Taipei Economic Office in Papua New Guinea," Chinese Foreign Ministry spokesperson Lin Jian also said during a press conference on July 16 that China highly commends the decision and it demonstrates once again that upholding the one-China principle is where international public opinion trends and the arc of history bends. 
Earlier in April, three African countries, Seychelles, Mauritius, and Madagascar revoked flight permit to cross their countries without prior warnings before regional leader Lai Ching-te's planned trip to Eswatini.
Zheng Jian, a professor at the Graduate Institute for Taiwan Studies at Xiamen University in East China's Fujian Province, told the Global Times that the Democratic Progressive Party (DPP) has pushed to internationalize the Taiwan question and expand "Taiwan independence" activities abroad, directly challenging the one-China principle widely upheld by the international community and provoking pushback.
International community, including African countries, would not choose to support "Taiwan independence" in their diplomatic decisions, and the more the separatists seek international space, the narrower their room for maneuver will become, the expert noted.
Against the backdrop of the one-China principle, attributing such entry restrictions solely to the Ebola-related policy would amount to misplaced wishful thinking, he said. In truth, DPP authorities should not ignore that all the setbacks Taiwan region faces internationally nowadays are brought about by "Taiwan independence" separatist activities, Zheng said.
A spokesperson from Taiwan Affairs Office of the State Council, Zhang Han, said during a press conference on July 22 that facts have proven time and again that the one-China principle is a basic norm governing international relations and a universal consensus among the international community. 
"Upholding the one-China principle aligns with righteousness and represents the irresistible historical trend. A just cause gains wide support, whereas an unjust one garners little backing. The DPP authorities stubbornly adhere to their separatist 'Taiwan independence' stance. Acting against the tide of history, they will only encounter repeated setbacks internationally," Zhang added.
。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。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。

B | 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。”除一种情况外，AI 模型给出的平均分普遍高于人工评分。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。

C | 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。

D | 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。

E | “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。

F | ”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。

G | 但这项研究表明，目前并不适合这么做。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。“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。