Category Archives: Tech Zone

#Keras tip: you actually need to specify batch size again when you run model.predict() after, otherwise it defaults to a batch size of 32. Depending on your model this could be too small, leading to incredibly slow inference. Common online examples do not seem to emphasize this at all.

由兩年前開始,人工智能已經可以做到換臉的效果。*隨著技術的成熟和普及,一些…不太見得光的用途亦開始出現。網上社群Reddit現時有一個群組專門把著名女星的臉配上成人影片之上。該社群還推出了一個應用程式把換臉程序大幅簡化,只要用戶找到影片,按幾個鈕就可以利用家中電腦的顯視卡製作影片,無須對背後的技術有任何瞭解。難怪網媒Motherboard的報導標題為「We Are Truly Fucked」,因為根本無法阻止有心入把別人當成對象製作影片。



We Are Truly Fucked: Everyone Is Making AI-Generated Fake Porn Now…/reddit-fake-porn-app-daisy-r…
Gender and Race Change on Your Selfie with Neural Nets…
Nothing is real: How German scientists control Putin’s face

Bitcoin transaction confirmation time today: 1,928 minutes.
That’s 1 day and 8 hours in case you’re counting.

E-Sports is still a taboo among many Hong Kong parents, who hold the outdated view that gaming is just waste of time with little reward.

(China just lost to France in the Quarter Finals of Overwatch World Cup. Team pictured below.)

真正靠人工智能賺大錢:顯視卡製造商NVIDIA股價今日首次突破200美元 。去年其股價才70美元,兩年前是30美元!
How A.I. makes you rich: NVIDIA reached $200 for the first time today. It was $70 one year ago and $30(!) two years ago.

網絡安全專家Mathy Vanhoef今日公開了一個影響*全部*Wi-Fi裝置的漏洞,這個漏洞讓駭客可以在沒有Wi-Fi密碼的情況下解密透過Wi-Fi傳遞的信息。雖然沒有證據顯示這個漏洞已被廣泛利用,但在漏洞公開後相信這只是時間問題。目前的最佳自保方法:

A protocol-level vulnerability allows an attacker to decrypt protected messages by attacking the four-way handshake of WPA1 and WPA2. Now more than ever: use HTTPS and VPN!

The sequential write speed of #SanDisk #Ultra #128GB #microSD is surprisingly fast, over twice that of the other 128GB card I have in hand, a Toshiba Exceria 128GB, in the Q32T1 benchmark. The figures seem too good to be true, but I have already repeated the tests 3 times for a total of 15 runs and the results still hold.

My students these days probably don’t know what these are.

Did you know that one of the pioneers in neural network was an econometrician? This was not an easy book to find, but it was well worth it. The book is a collection of articles written by Halbert White—as in *White* Standard Errors—that rigorously laid the theoretical foundation for the effectiveness of neural nets, including universal approximation, consistency and asymptotic normality of learned parameters. These results fill a much-needed gap for those with a background in statistics.