I think those are ignored.
cool. perfect. I thought so
Yes, this is a serious issue. For instance, YouTube’s recommendation algorithm tries to maximize amount of time people spend watching YouTube, and inadvertently learned that conspiracy theorists watch significantly more YouTube than people that trust other media sources. Thus, it disproportionately recommends conspiracy theories.
24 Likes
If you’re into spreadsheets, I expanded on Jeremy’s model and built this Excel model (linked to in blog post) on collaborative filtering a few months ago. It shows the gradient descent formulas (instead of relying on the built-in solver) and has drop-downs to change hyperparameters.
16 Likes
Second. Companies maximize profits, not social welfare. That is a problem.
1 Like
I guess YouTube is turning me into a conspiracy theorist and I’m not actually imagining it. I knew it.
4 Likes
Jeremy is restarting his computer and will restart the live stream once its back up
8 Likes
Did excel solver kill jupyter?
7 Likes
Google search. Paper from folks at Google.
https://arxiv.org/abs/1703.00397
We are starting with the simplest version, but will build this into a non-linear neural net next.
2 Likes
Is there a way to build a structured data model for time series data that will use several rows of data to predict the subsequent row? For example taking data from Monday-Thursday to predict Friday’s values for all columns in the dataset.
2 Likes
What is that artwork/screensaver?
1 Like
In China, there is a news app that centers around recommendation system. People love using it and it is making huge profits, but the top news pieces recommended are anything but good journalism; I would call them psychological weapons that exploit people’s cognitive biases and flaws.
5 Likes
I will post once the stream is back up.
6 Likes
19 Likes
Was the freeze due to running the excel solver?
3 Likes
How do you validate the model that used collaborative filtering? How do we know it is a good model?
2 Likes
Nope but due to running too many things at the same time. For example streaming software, multiple browsers with multiple tabs, excels, ppt and even Skype.
There was a widely reported-on story a few weeks ago when Amazon tried to use machine learning to expedite its hiring process and the model learned to bias against women. This was reported as “AI discriminated against women” and the like, but arguably what really happened was it uncovered historical bias against women in Amazon’s hiring practices.
9 Likes
Check out Knowhere for news that AI has made “neutral”, you can also select to see the different “sides” of the story if you actually want the biased versions.
1 Like