One of the big questions of our time is whether AI can improve AI without any human assistance. Here’s an X post from Yann LeCun, someone who is considered to have done useful work in AI:
He responds to the (admittedly absurd) reveal of the Tesla Roadster with “Cute demo, but totally impractical.”
Can we infer from this that Yann LeCun has never met or observed a fellow human? Most human purchases are, of course, impractical. Almost nobody needs a pavement-melting SUV. If “practical” sold cars, every car would be either a Honda Civic or a Toyota Sienna, depending on family size. (Our beloved Honda Odyssey would get the axe because, though it is more fun to drive than the Sienna, it uses more gasoline to accomplish a similar practical mission.)
If LeCun can be useful in AI without knowing the first thing about humans, should we infer that AI left alone will produce better AI?
In case the above tweet is memory-holed and with the bonus inclusion of a thoughtful response from Tesla’s CEO:
For us non-technology educated readers, what “useful work” has LeCun accomplished in SI?
I’m not sure that Yann LeCun has done that much recently, e.g., on the Transformer architecture that Google published in 2017 and that is the basis of all of the current wizardry.
https://amturing.acm.org/award_winners/lecun_6017366.cfm
talks about what made him such an important nerd. He worked on “CNNs” and perhaps that is what has made him such a passionate progressive Democrat.
—————- the most relevant paragraph
LeCun’s work at Bell Labs focused on the neural network architectures and learning algorithms. His most far-reaching contribution was a new approach, called the “convolutional neural network.” Many networks are designed to recognize visual patterns, but a simple learning model trained to respond to a feature in one location (say the top left of an image) would not respond to the same feature in a different location. The convolutional network is designed so that a filter or detector is swept across the grid of input values. As a result, higher level portions of the network would be alerted to the pattern wherever it occured in the image. This made training faster and reduced the overall size of networks, boosting their performance. This was an extension of LeCun’s earlier achievements, because convolutional networks rely on backpropagation techniques to train their hidden layers.
In short, LeCun showed that a neural network could figure out what features to look for, e.g., for image recognition, without humans telling it “look for edges”.
I guess another explanation for why he’s such a Trump-/Elon-hater is all of his years at NYU surrounded by the folks who would ultimately become the Queers for Palestine. Plus, everyone in academia wants Democrats to win and expand government funding for universities.
Since LeCun is French and a huge progressive it would be interesting to hear his explanation for the recent peaceful riots in Frogland. France already has everything that a progressive might dream of: super high taxes, even larger welfare state (percentage of GDP) than the U.S., universal health care, hardly any disgusting rich people (since the whole place is kind of poor and they have no tech industry, which is what has been making the U.S. rich lately), tons of low-skill migrants from all of the world’s most dysfunctional societies. Despite the place beina a progressive paradise, the progressive youth aren’t happy. How is that possible?
@Philip, You left out one important “feature” of France’s ultra-progressive model: the difficulty companies face when trying to fire or lay off employees. If someone isn’t producing, good luck getting rid of them.
Who are you calling a “frog”, Monsieur Phil?
Now, I don’t wanna talk to you no more, you empty headed animal food trough wiper!
“Yann LeCun provies”
You don’t even know how to spell words in your own inferior stolen language. You don’t frighten us, English-speaking American pig-dogs! Go and boil your bottom, son of a silly person.
Vive la France. 🇫🇷
Henri: Merci beaucoup! Réparé!
(I’m sorry about your death at age 55, supposedly due to “heart attack” but we all know that the cause was nuclear power or, perhaps, marrying a 25-year-old when you were 37.)
The first trainable CNN was created by Wei Zhang in 1988, before LeCun in 1989, but Wei Zhang called it a ‘shift-invariant pattern recognition neural network’ instead of a CNN.
https://people.idsia.ch/~juergen/who-invented-convolutional-neural-networks.html
Yeah, you’re supposed to say SI instead of AI this week, people. Maybe we should have that little talk.
As one in the permanent startup underclass, it is somewhat painful to see so much capital going to engagement projects instead of the desperate startups yearning to make next week’s payroll. Yet we realize the new economy is driven by engagement rather than solving problems.
It is clear that LeCun dislikes Musk and what Musk has accomplished. LeCun complains about the “compressed air” but ignores the car’s suction capability. The fact that the car could remain upside down before performing the flip is impressive in itself. It demonstrates what is technically possible, even if it is not something anyone need in real life.
Perhaps LeCun should turn his attention to the dresses that designers, especially those in Paris, showcase on the catwalk and tell us how useless those dress are. To that end, perhaps France should outlaw fashion shows because those are “haram”.
I fart in your general direction 💨
Thank You, philGPT.