Peter Norvig - The Unreasonable Effectiveness of Data
UBC Computer Science3 HN recommenders · 2012–2026
74K views · lecture · evergreen · Uploaded 2011-10-11
Peter Norvig's UBC distinguished lecture on how massive amounts of simple data, such as for language modeling, can yield understanding that hand-built models could not.
Recommended as a favorite tech talk showing how huge data availability changes how problems are approached.
2:25:52The spelled-out intro to neural networks and backpropagation: building micrograd
Andrej Karpathy
18:40But what is a neural network? | Deep learning chapter 1
3Blue1Brown
19:09A Better Default Colormap for Matplotlib | SciPy 2015 | Nathaniel Smith and Stéfan van der Walt
Enthought
58:51DON'T PANIC — Hans Rosling showing the facts about population
Gapminder Foundation
1:30:28AlphaGo - The Movie | Full award-winning documentary
Google DeepMind
27:14Transformers, the tech behind LLMs | Deep Learning Chapter 5
3Blue1Brown
51:22USENIX Security '18-Q: Why Do Keynote Speakers Keep Suggesting That Improving Security Is Possible?
USENIX
1:08:40Lecture 1 | Machine Learning (Stanford)
Stanford
5 comments from Hacker News
I am not sure if I disagree with the central thesis of the article, but the argumentation seems rather weak in many places, for example he says: Let's take the example of the number of people who walk into a store and buy a certain product. By studying 10 customers you can get an estimate of the probability of the purchase and predict sales; by studying 100 customers you get a better estimate. By analyzing 1000 customers you have a very good estimate. You could study a million customers, but the results are unlikely to be vastly different from the results of analyzing 1000 customers and probably not worth the expense. This makes sense only for very low-dimensional data sets, but most real-life problems are very high-dimensional and in high-dimensional spaces a million customers (or any other number of samples you might reasonably expect to gather) might still very well be only a small sample. So, there are situations where because of various constraints you will never be able to get to this level where more data stops yielding better results. You can watch a nice presentation from Peter Norvig about the unreasonable effectiveness of data: [VIDEO] Or read about the well-known curse of dimensionality: [LINK]
If you're looking for more examples and information about using simple algorithms with lots of data, I recommend Peter Norvig's "The Unreasonable Effectiveness of Data": [VIDEO]
This is old news: Peter Norvig, The Unreasonable Effectiveness of Data (2011): [VIDEO] and a previous article from 2009: [LINK]
Peter Norvig on the "Unreasonable Effectiveness of Data" [VIDEO] I think it is so easy for us to discuss the impact of big data and quickly get into the weeds, but I think in this talk Norvig does an especially great job in making you truly appreciate the seismic impact that the availability of massive quantities of data can have on your way to think about problems. This is one of the first things I ever saw of him, and I've been in love ever since.
That happens often with domain names, but then you get expertsexchange.com, penisland.net, whorepresents.com, therapistfinder.com, a Dutch pre-match analysis site voorspel.nl, or a site about the game overspel.nl. Peter Norvig - The Unreasonable Effectiveness of Data [VIDEO]