谷歌 Gemini 发布后遭质疑：测试标准有失偏颇，效果视频疑似剪辑
谷歌 Gemini 在凌晨发布后吸引了巨大的关注。然而，有网友指出，MMLU测试中，Gemini 结果下面灰色小字标称 CoT@32，展开来代表使用了思维链提示技巧、尝试了32次选最好结果。而作为对比的GPT-4，却是无提示词技巧、只尝试5次，这个标准下Gemini Ultra其实并不如GPT-4。
此外，机器学习讲师 Santiago Valdarrama 认为 Gemini 的演示视频展示的是精心挑选的好结果，而且不是实时录制而是剪辑的。这些质疑引发了关于 Gemini 测试标准和演示视频真实性的讨论。
在这种情况下，我们建议读者保持谨慎态度，并对 Gemini 的性能进行进一步测试。同时，我们也期待谷歌能够对这些问题作出回应。
总之，谷歌 Gemini 虽然在发布后吸引了巨大关注，但也面临着一些质疑。我们将继续关注这一事件的发展，并及时更新相关信息。敬请关注。
英语如下：
====
“News Headline: Google Gemini Release Raises Concerns: Testing====
“News Headline: Google Gemini Release Raises Concerns: Testing Standards Biased, Effect Video Suspected of Editing
Keywords: Google Gemini, Testing Standards, Effectiveness Doubts
News Content: Google Gemini Released and Faces Doubts: Testing Standards Are Biased, the Effect Video Is Suspected of Editing
Google Gemini attracted huge attention after its release in the early morning. However, some netizens pointed out that in the MMLU test, under the gray small print labeled CoT@32 on Gemini’s result, it means using chain-of-thought hints and trying 32 times to choose the best result. As a comparison, GPT-4 has no hint words and only tries 5 times. Under this standard, Gemini Ultra is not as good as GPT-4.
In addition, machine learning lecturer Santiago Valdarrama believes that the demonstration video of Gemini shows carefully selected good results, and it is not recorded in real time but edited. These doubts have triggered discussions about the testing standards and authenticity of the demonstration video of Gemini.
In this situation, we suggest readers to remain cautious and further test the performance of Gemini. At the same time, we also look forward to Google’s response to these issues.
In summary, although Google Gemini attracted huge attention after its release, it also faces some doubts. We will continue to follow the development of this event and update relevant information in a timely manner. Please stay tuned.”
【来源】https://www.qbitai.com/2023/12/104425.html
Views: 7