A portfolio project for undergrads? There's a 300 lines of code keras.io example that does exactly this (with better performance). Anyone with some Python experience could read it and modify it.
Machine learning just keeps getting easier and more accessible.
One thing I want to emphasize: ease-of-use is obviously beneficial to those with less expertise, who go from 0 to 1. But the people who gain the most from it are actually the experts, who are now able to dream bigger and move faster. Experts go from 1 to 10.
"My expertise is devalued, undergrads can reimplement my PhD thesis in 30 minutes 😢" is the wrong mindset.
It's actually: "my expertise is being multiplied, I can now achieve dramatically more ambitious milestones while leveraging more enjoyable and productive workflows 🚀🚀🚀"
"Oh no, better tools! They're helping the kids catch up to me!"
vs
"Yay, better tools! Can't wait to see how far I can go with them!"
• • •
Missing some Tweet in this thread? You can try to
force a refresh
Here's a trivial example to illustrate the difference between pattern recognition and reasoning and it impacts behavior generation: let's say you encounter, for the very first time, a glass door with ⅃⅃Uꟼ written on it.
Pattern recognition: nearest neighbor is "PULL", I've learned to associate that with pulling the door. I pull.
Reasoning: that's a mirror image of "PULL". Must be written on the other side. The door will open by pulling towards the other side, i.e. pushing from this side. I push
Some apparently read this as "all recommender tech is bad & dangerous" which is not it at all.
The takeaway is that the information you consume matters and your attention is precious, so you should have very high standards for what's allowed to download thoughts into your brain.
You wouldn't eat random crap handed to you by a greasy robot on the street. Being deliberate about what you eat is critical to your health.
Well, being deliberate about the information you consume is critical to your mental well-being.
That doesn't mean that you should not eat anything or that you should never go to restaurants. It just mean you should have standards and you should be mindful of what you eat.
I hear people saying "new housing is getting bought up as an investment vehicle, building more just helps the rich". Of course this is factually incorrect, but even more importantly, *in a world where enough housing gets built, housing is a depreciating asset, not an investment*.
When there is sufficiently supply, the price of a house / apartment goes down over time as it gets older (thus less attractive) and needs repairs. Eventually you have to tear it down and rebuild. Just like a factory. Or a car.
The price of housing only goes up because of artificial constraints that prevent supply from meeting demand. Housing works as an investment because we don't build fast enough / dense enough.
People get scammed not so much because they can't analyze a scheme on an intellectual level, but because of their personality or emotional state. Smart are often *more likely* to get scammed, because they're more confident in their own judgement.
"Is it a scam?" A checklist.
1. Is there a real product/service involved, with an actual business/consumer need? What does the thing look like if you consider it from 1st principles, rather than from the lens of "smart people are into it, must be legit" or "even my neighbor is making money, must be legit"?
(scams will often try to hide their lack of practical foundations by saying they represent a "new paradigm", like "the internet in the 90s" -- everything that involves tech and hucksters must be like the internet, nevermind what the internet was actually useful for back then)
The tech industry's successive waves of fascination with get-rich-schemes powered by FOMO ("my idiot neighbor is getting rich, I have to get on this") actually serve a useful social purpose.
It compartmentalizes the negative impact of a certain kind of people/activity.
By default, the get-rich-quick folks go into what's hot. And you wouldn't want them to be founding AI startups promising the Singularity in 5 years, or biotech startups promising 100s of blood test results from a single finger prick. The fallout would be devastating.
Well-isolated tech grifter ecosystems work as a magnet that shields productive areas of research and product development from conmen, their deleterious mentality, and their long-term impact on public confidence.
If the 3rd AI winter never arrives, it's thanks to them.
Use "mixed_float16" as the policy value on GPU, and "mixed_bfloat16" on TPU.
You can also configure the dtype policy on a per-layer basis using the `dtype` layer constructor argument. Useful to keep some layers running in float32 when necessary!
In our example, we kept the data augmentation stage in float32 (created before setting the global policy) since it's meant to be run on CPU as part of the TF data pipeline.