If you don't mind me asking, what social media are your family members using? Are their friends on that platform too?
2,242 karma · joined February 8, 2015
Yann Lecun himself (as of NIPS2016) was pretty critical of probability-based metrics, as those have strong dependence to the choice of model (e.g. if the model is poor the log-probability is meaningless).
In GANs, the critic and the generator are trained w.r.t. each other, reaching some kind of equilibrium. A recent proposal that seems to be "ok" for evaluating GANs was proposed by https://arxiv.org/abs/1705.05263, which is to train a separate critic on the generator, for use in evaluation (the generator never sees gradient information from this critic). This evaluation critic approximates the Wasserstein distance. One could imagine actually training the independent critic on a validation set of images not seen by the training set.
Though he doesn't mention who the "friend" was, is it the much-beloved (by Brown CS dept) Andy van Dam (who is credited as the co-designer of HES)?
Oh, no, sir. This wasn't built by a magician. This was built by a wizard. A man who can actually do what magicians pretend to do.
He does things that I wouldn't have dreamed possible with fragment shaders and implicit geometry. https://www.youtube.com/watch?v=yMpG6qEb8js
IQ is a wizard and really inspired my interest in programming and math.
- Is she conscious at all? Is she conscious just like we are? Is she conscious in some kind of weird alien way. You never really know.
- we’re pretty close to human brain scale computing already in the world’s fastest super computers. We’ll get there within the next couple of years. But that doesn’t mean to say we know how to build human level intelligence. That’s an altogether different thing. And also there’s controversy about how you make that calculation as well. I mean do you, what do you count? Do you count a neuron? How do you count the computational power of one neuron? Or one synapse? And some people, you know, it may be that some of the immense complexity in the synapse is functionally irrelevant.
- They’re just the process of turning the raw waveform into text. So that’s been cracked. But then again, real understanding of the words, that’s a whole other story. And with all today’s personal assistants, you know it can be cool and they’re gonna get better and better, they’re still way off displaying any genuine understanding of the words that are being used.
- But let’s suppose that it was a world where we were minded to put Asimov’s Laws into Ava. Well maybe Ava might reason that she is human. You know what is the difference between herself and a human? And maybe she would reason that she shouldn’t allow herself to come to harm. And therefore she was justified in what she was doing. Who knows?
- Of course there are people who want to build autonomous weapons and all kinds of things like that and you might say to yourself well I would very much like it if somebody was trying to pay attention to things a bit like Asimov’s Laws and say well you know, you shouldn’t build a robot that is capable of killing people. But that’s a law that the designers, or that would be a principle, if we were to have it, that the designers and engineers would be exercising, not one that the robot itself was exercising. So that’s the sense in which it’s not relevant toady because we don’t know how to today make an AI that is capable of even comprehending those laws.
- Nathan’s criticizing Caleb over going with his like gut reaction and his ego and not in like if he were to think through every logical possibility for every action he would never do anything. Which is kind of like directly against all these laws. Ava would never do anything if she could harm someone possibly down the road by you know burning fossil fuel, by being in a helicopter.
(My own commentary) This last point may be particularly relevant to researchers in Deep RL. Often we formulate off-policy RL problems as computing value functions that represent expected reward over optimal policies until the end of time. But maybe long-horizon reasoning is too computationally expensive (requiring agents to solve every trolley problem in its future and model the effect of burning fossil fuel on the environments), so some kind of rational ignorance is actually optimal in a compute-resource-restricted regime.
https://www.nytimes.com/2015/10/18/books/review/the-song-mac...
1 - 200/6158 = 97% is indeed a pretty suspicious coincidence though. I would assume/hope that they've shuffled a big dataset of recorded heart events (like the image in the TC article) and that the 200 people diagnosed with paroxysmal atrial fibrillation only rarely experience AF, so the number of true positives is probably far smaller than 1% of the dataset.
In Machine Learning, the standard expectation of research papers is to empirically demonstrate the increased statistical power of the model over benchmarks, and I wish the authors demonstrated that their model aligns with historical reality.
I attended a talk recently by Aswath Damodaran where he explained that there are two types of investors, most often lying in disjoint sets: 1) number crunchers and 2) story tellers. Perhaps the success of tech companies (e.g. AMZN, GOOG, FB) can be explained in aggregate via the Wright model, but my feeling is that the majority of 'optimal portfolio distortion' these companies caused is mostly idiosyncratic to those specific companies, e.g. the "story tellers" have more information in the long-term behavior of a company. Consequently, aggregate modeling captures statistical correlation rather than causation, which can still make money but is not robust to long tails.
I think number crunching is only useful for short horizon portfolio selection.
- Nathan, Ex Machina (2015)
Imagine if this discovery were made by some undergraduate student who had little experience in the traditions of how ML benchmark experiments are done, or was just starting out her ML career. Would we be just as critical?
As a researcher, I like seeing shorter communications like these, as it illuminates the thinking process of the researcher. Read ML papers for the ideas, not the results :)
I personally don't mind blog posts that have a bit of hyped-up publicity. It's thanks to groups like DeepMind and OpenAI that have captured public imagination on the subject and accelerated such interest in prospective students in studying ML + AI + robotics. If the hype is indeed unjustified, then it'll become irrelevant in the long-term. One caveat is that researchers should be very careful to not mislead reporters who are looking for the next "killer robots" story. But that doesn't really apply here.
Regardless of what one thinks regarding AI displacing jobs, there's been a remarkable growth spurt of technology and that begs lots of uncertainty about the future. When things get uncertain, people will speculate more, push their opinions on others, etc.
Sorry I'm not answering your question directly, but hopefully it suggests why we might be seeing all these "what's going to happen with AI" articles. You can bet there will be lots of opinions coming soon on "what should we do about it", but these are harder to tease out.
The data gathered using aforementioned robot arms is available here: https://sites.google.com/site/brainrobotdata/home
Google (specifically Google Brain Robotics) is continuing research on this. Exciting time to be alive.
I don't think articles like "Study: Immigrants Founded 51% of U.S. Billion-Dollar Startups" are all about anti-Trump, it's more like pointing out to the American public that immigrants are not a tax on the rest of society.