AI in use | Microsoft CEO's crazy Amway AI game tortured me thousands of times

Machine Power Report

Editor: Yang Wen

The wave of artificial intelligence represented by large models and AIGC has quietly changed the way we live and work, but Most people still don't know how to use it.

Therefore, we have launched the "AI in Use" column to introduce the use of AI in detail through intuitive, interesting and concise artificial intelligence use cases and stimulate everyone's thinking.

We also welcome readers to submit innovative use cases that they have personally practiced.

Oh my god, the AI ​​is really awesome.

Recently, it’s difficult to distinguish the authenticity of AI-generated pictures, which has caused quite a stir. (For details, please go to: AI in use | Become an AI beauty in three steps, and be beaten back to your original shape by AI in a second)

In addition to the popular AI Google lady, there are all kinds of AI beauty popping up on social platforms The photos of people generated by FLUX are all lifelike. (To view the details of FLUX, please go to: Actual test of the latest Vincentian artifact FLUX: Midjourney? It’s too early)

(Click on the picture to enlarge the view)

Taking advantage of this wave of popularity, Microsoft President Brad Smith released an AI testing game on the X platform - Real or Not?

The gameplay is very simple. Each round of the game has 15 pictures, and players decide which one is AI-generated and which one is real.

We tried it out and guessed 6 out of 15 pictures correctly. The correct rate was less than half. I really don’t know whether I am blind or the AI ​​is too weird.

-1-

yo! This world of real and fake

Recently, Microsoft President Brad Smith posted a photo on the X platform with the following text: "Can you tell whether this picture is real or fake? Take this test Test your AI recognition skills and share your scores to help fight against the abuse of Deepfake technology."

At the same time, he also gave a link to this mini game:

https://www.realornotquiz. .com/

How real are the pictures here?

Look at this lady, the crow’s feet that naturally wrinkle at the corners of her eyes when she smiles, the nasolabial folds at the corners of her mouth, her light brown eyes, hair, and even the faintly visible spots are all very real, but the result——

She is actually an AI.

Look at these four small yellow flowers again. The picture quality is too clear. The petals in the lower left corner seem a little distorted. They look particularly like the "masterpiece" of AI convulsions, but -

This is real of!

I really like this old 6! A stumble at the start.

A few more pictures for everyone to guess:

(Click on the picture to enlarge it)

Yes, the above All four photos are real.

After our keen eyes, we finally correctly identified 40% of the pictures.

After many experiments and inspiration from netizens, we discovered a bug in AI-generated images.

Look at the details first, especially whether the fingers are deformed, whether the shadow is distorted, etc.

For example, in this picture of multiple people playing guitar, if you look closely, you will find that the person playing guitar on the sofa on the left has an extra hand out of thin air.

Another example is this silhouette of the seaside at sunset. At first glance, there is nothing wrong with it, but looking at the shadow of the man, his head looks like a robot, so I can tell that it was generated by AI.

Second, look at the picture quality and skin texture.

If the image quality is too clear, the skin is so delicate that pores cannot be seen, and the forehead and cheeks still appear shiny, this is most likely generated by AI.

（點圖可放大查看）

第三看是否符合實體世界邏輯。

例如消防員滅火時噴灑的乾粉滅火器，怎麼可能像下過雪一樣，堆得到處都是？

一輛生鏽破舊的坦克，大機率不會丟棄在遊樂園附近。

-2-

你到底是不是人?

無獨有偶。

去年還火了一款號稱「社群圖靈測試」的AI 小遊戲－Human or not?

官網連結：https://app.humanornot.ai/

此遊戲綜合了Jurassic -2、GPT-4、Claude 和Cohere 四個大語言模型。

遊戲規則：系統在線上分配一個夥伴（partner），你們只被允許限時聊 2 分鐘。然後，系統讓你回答螢幕對面的這個 partner 究竟是人類（Human）還是 AI 機器人（AI Bot）。

大家猜猜對方是人類還是機器人？

這英語表達、這傲嬌的語氣，多像個真實的人類，但…… 它是個機器人。

再來一個：

聊完一個回合，在遊戲時間未耗盡的情況下，對方突然離開，毫無疑問，這就是人類。

而這種有問必答，禮貌得體的，則是 AI。

經過多番較量，我們摸索出一個經驗，回答簡短的不一定是 AI，但回答得長篇大論的，大概率是 AI，因為 20 秒根本來不及打太多字。 （每次回答都有時間限制，20 秒）

此前，《每日郵報》的記者也進行了相關測試，在 10 次對話中，有 6 次猜對。報導稱，雖然 AI 偶爾會出現糟糕的語法，但它確實成功地複製了人類的語言，讓「Human or not？」的遊戲變得極其困難。

早在 70 多年前，電腦科學和密碼學先驅艾倫・麥席森・圖靈提出「圖靈測試」這個概念。

他認為，測試者與被測試者（一個人和一台機器）在隔開的情況下，透過一些裝置（如鍵盤）向被測試者隨意提問。經過多次測試後，如果機器讓每位參與者做出平均超過 30% 的誤判，那麼這台機器就通過了測試，並被認為具有人類智慧。

AI 發展到今天這個地步，社交媒體上早已充斥著大量真假難辨的圖片、視頻，甚至聊天時都不知道對面是真人還是 AI。

我們在體驗到新奇和有趣的同時，也會有些恍惚和「小小的恐懼」，《黑鏡》降臨的世界究竟如何。 。 。

工具連結－

Real or not 連結：

https://www.realornotquiz.com/ :// app.humanornot.ai/

以後我們會帶來更多好玩的AI用例，也歡迎大家進群交流。

The above is the detailed content of AI in use | Microsoft CEO's crazy Amway AI game tortured me thousands of times. For more information, please follow other related articles on the PHP Chinese website!