Ang Li, co-founder of Simular, reflected on his experiences at Google DeepMind, noting early skepticism regarding machine learning's efficacy among software engineers. Between 2017 and 2019, many believed that machine learning rarely worked in production settings. Li cited attempts to implement the AlphaGo system to boost Google Ads revenue, which instead resulted in decreased revenue. He emphasized that machine learning relies on static datasets, a premise that fails in dynamic environments like YouTube and digital ads where data is continually changing. Three years post-ChatGPT, machine learning still faces significant limitations in accuracy.
Machine learning methods are based on statistics, and they assume a static dataset. But in the real world, this assumption doesn't hold because data keeps changing.
The Google Ads team asked the DeepMind crew to apply its AlphaGo system to improve ad revenue. However, it actually dropped the revenue, highlighting the complexity of real-world systems.
Collection
[
|
...
]