goldemerald

Great podcast Lex!

I am intrigued about the discussion on explainable AI. How do you feel about the quality of the current XAI research? What do you think are the most important directions for that field? And what do you see it looking like in 3-5 years?

My goal is to one day have Schmidhuber angrily claim that my research was done by him in the 90s like what happened to Ian Goodfellow [0].

[0] https://www.reddit.com/r/MachineLearning/comments/5go4sa/n_w...

You can enter scroll mode with ctrl+b+[ then you can mouse wheel, PageUp, etc.

To exit just press q or ctrl+c

Does your work integrate pretrained LMs like BERT or GPT2?

Specifically, here's the relevant text from the paper:

"...extended overnight fasting periods might increase the risk of gallstone disease (Sichieri et al., 1991)"

Is there any comparison of training speed of a neural net for one of these chips and a typical one? I'd be interested to see how long it takes to train an imagenet classifier on one of these compared to other hardware.

This stuff is great! Will example training files be uploaded for GLUE/SQAUD/etc like on the HuggingFace's implementation?

There is an ethic commitee (the IRB) at every university that reviews all academic research involving people. In order to conduct a user study, researchers are required to compensate participants fairly. The reason you can pay MTurkers "far less" is that a study may only take 5 minutes, and researchers typically pay $10/hour. People won't participate in a study unless the minimum payment is decent. I can't imagine many people would want to take the time out of their day to show up at a lab, work on a computer task for 10 minutes, and get paid $2.

This article isn't too surprising. Modern studies only find a small positive connection between IQ and lifetime success (0.25 correlation). Even so, there is no single element predictor that can better measure success compared to IQ. I would hope that by 2059, social scientists can find an accurate way to determine how being rich contributes to "winning".

I appreciate your desire for a better hackernews community, but I disagree with your assessment of my comment. The article is clearly for marketing, not quality content. The original poster is indeed a member of the company. That's not to say company articles are all bad, but the merit of this article is what I based my comment on. But to be fair, what I consider "bile spewing" is based on reddit quality comments, so perhaps I should rethink my commenting style.

That's a funny company. They patented an implementation of a Bayesian Network and call it something new and amazing.

Not that it matters, as I'm certain other companies use more modern techniques (deep learning) for "outfit advice".

My apologies, I wasn't stating his thesis in full. "Poorly defined" in the scientific sense. And to address your concern: a title should not be a thesis, thesis' of articles can be implied, and other commenters to not appear to be discussing the content of this precise article.

I typically don't like to read into the motivations of articles (or people's actions for that matter) and I take their words as an independent object. Given the article has a logical flow with plenty of high quality citations, I'd say his thesis of "willpower is ill-defined" is a valid statement.

EDIT: The author is an Assistant Professor of Clinical Psychiatry at Columbia University, whether or not that affects your view of the article is your prerogative.

I disagree with your assessment of the article. The Victorian definition of willpower is the same one my friends/family and I use today. Now that I have read the article I will be more mindful of how I use the term "willpower" compared to other concepts (like the cognitive strategies).

The author makes a great case for how we use the term willpower is different scenarios to mean entirely different concepts. While the marshmellow experiment willpower is similar to interpersonal bargaining willpower, the difference is important for defining scientific vocabulary. The author is not talking about colloquial use of willpower. And these two separate use cases of the term need to be rectified.

Granted, the author doesn't discuss solutions. But it seems to me a discussion of solutions would be better as another article.

I feel like the other commenters here are missing the point of the article. The current scientific understanding of motivation and willpower is severely lacking. The author is pointing out this fact. And while he does not have a thorough discussion about a solution, it doesn't take away from his thesis: willpower is poorly defined.

I can assure you there is a lot of interest in the Machine Learning field to make deep fakes for music. The problem is deep fakes are not time-series based. Right now a video is treated as a bunch of independent images, whereas you couldn't do that for music/text as there are long term dependencies between time steps.

More specifically, there is a recent paper from ICLR about audio GANs (I haven't read it) https://openreview.net/forum?id=H1xQVn09FX. IMO, this tech should come out (even in app form) in the next 2-4 years.

I am one of those people who listens/watches all media at 2x speed. It takes some time to get used to and is nearly impossible without headphones. Some people do talk quickly enough that I can't listen to them while doing any complex activity (e.g. Mike Duncan's revolutions podcast). But most other podcasts aren't scripted, so 2x is easy enough for speakers like Joe Rogan and Dan Carlin.