I agree. I think a big part of this problem is that smaller companies usually cannot afford AI research. I would even go as far as to say there are more AI companies than capable AI researchers, and this causes a large number of faux-AI companies poisoning the AI branding.
We are not capable of recreating human level intelligence yet, but our modern algorithms had become magnitudes better at generalization and sample efficiency. And this trend is not showing any signs of slowing down.
Take PPO for example (powers the OpenAI 5 dota agent), the same algorithms can be used for robotic arms as it does with video games. Two completely different domains of tasks now generalizable under one algorithm. That to me is a solid step towards more general AI.
It appears they are using an algorithmic process to perform the heartbeat trading, and they are patentable[0] as long as they are not illegal. Tax loopholes are not illegal but very unethical, I think the policy makers share equal if not more blame for this.
This is such a cool concept, I really wish these guys would succeed. Though last time I heard they gave up developing their own rocket (as a second stage to space). Anyone who's in the loop, does Stratolaunch currently have any potential rockets lined up?
Occam's razor applies here. Your theory is that she had 1. intended to assassinate Assange, 2. attempted to cover it as a joke and then 3. further covering it up by implying she doesn't remember this statement.
Let me offer you another theory: she's almost certain that she didn't say it, but she couldn't remember every word she's said.
Which theory do you think is more likely?
Now I am not saying you are wrong, but you'd need external evidence to support your theory, until then, Occam's rule apply. If you agree that my theory is more probable but do not like it, then it's possible your current belief is skewed by your prior belief.
Sure theoretical advances may be evolutionary, but they resulted in exponential reduction of parameter space and sample complexity. These advances outpace Moore's law by a large margin.
Hardware served as a catalyst, but it was not a necessity.
Less than four years ago our state-of-the-art RL system (DQN) could only beat some Atari games. Now we can almost beat best human player in SC. That to me is very impressive.
Currently visiting China for Chinese New Year, and can't help but to notice the anti-US anti-Apple sentiment that arose recently. Although anecdotal, I think the nationalistic sentiment has a lot to do with the dip in sales.
To put in perspective, most Chinese netizens are generally pretty critical of the Chinese gov. Chinese with higher educations usually look up to western culture and products. However after the Huawei arrest incident, the Chinese netizens were pissed at the American government, I've never seen such anger towards the US gov, and huge waves of boycotts began. They think that American gov is bullying a Chinese company into obeying US law even outside US, and their response is to boycott American products.
During my two weeks here I've talked to a dozen random people at bars, friends gatherings and on the plane. Normally US-China politics never come up, but it has come up in almost every single conversation this trip. Most of these people mention that they recent switched to a Chinese phone or their next phone will be a domestic phone.
Sure. I don't have any holistic survey to prove my point, but an example of recent progress in terms of sample efficiency is this paper[0]. Derivatives of this paper have been used to solve Sudoku[1], Starcraft II[2] and more [3]. This paper enabled more efficient use of data by creating a probabilistic graphical model between logical sets.
The sample efficiency of ML systems are increasing rapidly in the theoretical realm, though the available data are growing magnitudes faster than the progress in ML sample efficiency. The article has a valid point: It's far cheaper to hoard data than to invent a more efficient ML system, thus we will see people race towards data rather than technical complexity.
Are you suggesting distributed NN training through user's browsers? This is a very sketchy area along the lines of browser based bitcoin mining. You don't want to run any computation heavy code on client side without user's knowledge & approval.
Though if you only meant performing inference in user's browsers, then the challenge would be to find a way for TF/pytorch pre-trained models to port and perform accurately in js. If successfully done, I can see some use cases here.
There is currently no NN learning algorithm that can handle massively paralleled training. We can use some simple fixes such as mean gradient but they have severe limitations, and their limitation grows as you scale up. Currently even with a handful paralleled GPU training, the gradient computation needs to wait for all GPU batch to complete and then return to CPU before the next batch, so your idea of infinite scaling is just a pipe dream at the moment. Sure Python can't be run on a million pages, but heck, no NN architecture can even handle more than a dozen paralleled computation in a stable manner yet.
JS/V8 may beat Python in general operations but definitely not in math specialized libraries. A lot of python packages support OpenBLAS and AVX/FMA by default which gives huge boosts to math/tensor operations. Correct me if I'm wrong, I don't think the js ecosystem is quite there yet except for experimental projects.