I only post when I think I can somehow add value to the discussion, at least value for some people who might read the comment. Obviously some people with in-depth knowledge of the paper and field in general will probably already be aware of the link I posted (for example), but others might find it an interesting avenue to explore. I know I often come to HN comments for this reason: a topic seems interesting, I want to learn of related stuff and/or get opinions from people with deeper expertise.
So even if it's only for "future reference", it might actually bring value _for some_. At least that's what I tell myself :)
But the whole idea of figuring out first principles and "fundamental" design choices is quite appealing. They have a section saying that a big goal here would be automatic design of optimal data structures. They talk of machine learning techniques (they mention Bayesian optimization and reinforcement learning).
That seems a very interesting research direction. In the same vein there's this talk by Jeff Dean about how they use ML at Google to replace all sorts of heuristics in data structure design (bloom filters etc.) with machine learning to optimize performance, though from what I recall it doesn't automatically change the algorithm itself:
It's a theme I like to read about (I mean "practical issues around ML in production settings"). I find company blogs and some research publications are great resources. Examples:
and many more. Google also publishes papers on various engineering practices obviously, some ML-related, but I can't find a blog where they focus on that specifically.
Also it's not "to keep up to date", but there's a great paper (from Google) that's often cited:
It talks about issues you face over the long run (I've experienced some of those). It
also provides interesting pointers for further reading, e.g. about "pipeline jungles".
If others have pointers, I'm curious to hear about them as well.
When I saw what you were doing I thought "oh so you point your phone at the sky and it shows commercial planes around you as dots over the horizon" (or something). Then I saw the living room visualization and I realized that's not what you meant by AR. That could be a nice feature though (just a suggestion).
I just tried this and Alt-l seems to move to the line above (equiv. to Esc+l), Alt-J the line below, and in gvim it produces some accented characters for me. It would have been a good combination for finger movement though.
I too hated the need to press <Esc> all the time when I started with vi. It sounded as bad for wrists as the Ctrl key for Emacs.
The trick I've found (I don't remember where) and used ever since is to map the "kj" sequence to <Esc>, since it's otherwise unlikely in normal text. It's just a simple instruction in the vim config:
In my case it's just discomfort but I'm curious as to why constraints around the waist don't get discussed more in articles about posture. For me it's one of those things that annoy all day long, hovering slightly below awareness level.
You jest but I tend to think it'd be much better for posture and breathing if pants with more flexibility and wiggle room were in fashion for men. E.g. sweatpants. I had a teacher at uni who wore those and I thought he looked rather funny, but in retrospect I can empathize.
I have a very similar experience with sit-stand setups. I have one at home and had one at work.
I stand up for routine tasks (email etc.) and sit down for deeper work. I find standing unconsciously restricts breathing (core muscles contract) so makes me exhausted more rapidly. I rarely work standing for more than an hour.
Reading comments here it seems a very common experience.
I think no matter how many studies on the subject there will be, personal experience dictates how we'll do it. Like you I'm just glad the option now is now socially acceptable to be a bit more mobile when doing desk work.
You're right about that. It might be that I suffer from a "person B" problem :)
However I still think that the "short-lived" nature of discussions here is a relatively independent problem. I think integrating a way to "subscribe to a story" with email alerts on HN itself (rather than independent apps like hnreplies.com), and _displaying_ the number of people who did subscribe, would pave the way for longer-lived discussions. By seeing the number of people who did "subscribe to the story" you'd know whether or not it's still worth posting a comment even though it's been a day or two.
Maybe that's the reason forums have longer-lived discussions. You can easily get alerts when there are new posts in a discussion where you've posted.
I don't often comment on HN and this (the short-timed nature of discussions) is one of the main reasons. The other main reason is that I have a high threshold for considering my participation is worthy of other people's attention.
More often then not, when that threshold is crossed, I realize I'm too late and my post will just sit there, unread.
Aren't there others who think it'd be great to have a "sticky" or "follow" button, in essence signalling "I'm going to follow this discussion for a few days / weeks, because I'm genuinely interested"?
Also if there was a way to easily "suscribe" to a thread (email alerts) that'd help. (Someone mentioned hnreplies.com in a sibling comment, never tried it, but I'll check it out)
That's what I don't get. Publishers have all the incentive to get behind a solution like this, but somehow I never see a "send a tip" or Flattr button anywhere (maybe I'm not looking?).
I just wish a bunch of well-known publishers would get together and agree on _one_ solution that they'd all add to their site.
It's just a matter of adding a widget on their article pages. They already have large readership. It could kick things off and maybe get over that network effect cold-start issue.
They could add a nice bounty like "make a first payment and don't see ads on our site for a month". Or similar.
A solution like this could be a good candidate because it's neutral. With Flattr or similar you're encouraging that particular middleman. However I think to inspire confidence in publishers they'd need to prove that this solution will have strong support for years to come (which is always unsure with open source projects).
EDIT: To be clear I'm mostly thinking of micropayments to news sites. The strategy is less clear when talking about electronic commerce in general.
which states that Montreal startups (by their definition) employ around 10,000 people. Of course those behind that study - local "startup scene" people and governmental instances - would have some incentive to inflate the numbers, but still, it seems larger than what you assume (1% of 66000 ~= 600 - 700 jobs). It's not Bombardier, but it's not negligible either.
I'm not sure what to say about your larger point, because I'm not sure if you're advocating something in particular. Are you saying it would be better if Facebook, Google etc. didn't open labs here, e.g. to avoid competition for local AI specialists? Or maybe you're just drawing attention to that drawback. In any case, I'd just say I'd hope it's not a zero-sum game. With better opportunities, the pool enlarges because 1) people come from abroad and 2) it makes sense for local people to specialize (e.g. pursue a PhD) in that direction.
Well to me it does seem, at least, as though the recent surge in public and private AI investments (Microsoft, Google, now Facebook) would have a positive local impact for AI researchers here. I mean with enough momentum it becomes self-sustaining with positive feedback loops: researchers have multiple local opportunities, so it makes sense to stay here in Montreal long-term, companies (even though they only need AI "as a component", as you say) have strong incentives to establish AI labs here, AI grad students have local job offers and don't get told they'd better leave for better career prospects, etc.
It's a bit early to conclude concerning the long-term "impact on the local economy" as you mention, but I guess it can't hurt to shine internationally in one more respect.
EDIT: I'd add also that there are quite a few AI-centered startups, where AI is the forefront and not just a component, e.g. look at this:
That article is pretty light on details. I wonder if he pointed towards a specific form of unsupervised learning.
Anyway it's pretty funny in light of an intro I remembered from one of his old papers:
"It would be truly wonderful if randomly connected neural networks could turn themselves into useful computing devices by using some simple rule to modify the strength of synapses. This was the hope that lay behind the original Hebb learning rule and it is the vision that has driven neural network modelers for half a century. Initially, researchers tried simulating various rules to see what would happen. After a decade or two of messing around, researchers realized that there was a much better way to explore the space of possible learning rules: First write down an objective function [...] and then use elementary calculus to derive a learning rule that will improve the objective function." [1]
ie. backprop
So actually backprop was the solution to all that initial "messing around" with unsupervised rules. Though of course to be fair (if I understand correctly) those rules had very little to do with modern "unsupervised learning" methods (e.g. autoencoders, which still rely on backprop or similar optimization).
I built something very similar too. My setup is to have two screens, one for standing, one for sitting (and two set of input devices). I close one screen when I'm using the other and I mirror the image on both screens.
I have another setup with an Ergotron Workfit (can be lifted up and down) but I find that the position is more adjustable with my mirror-screen version. With the Ergotron the position of the hands relative to screen is fixed so eye-screen distance can't be adjusted as well as I'd like. Also screen height doesn't go as high as I'd like.
Basically you take notes like you normally would and the "diffs" of your notes become the elements you will review using spaced-repetition style intervals.
Drawbacks:
- It's not exactly user friendly in its current implementation.
- There isn't any way to give yourself a grade based on how well you remembered a piece of information.
- You must take care to write notes in small blurbs before running the "diff" command, otherwise I end up with large chunks of text to revise and I won't review properly.
- The lack of question-answer mechanism pretty much bypasses the whole "active recall" principle, or whatever that was called.
https://news.ycombinator.com/item?id=16230668
I only post when I think I can somehow add value to the discussion, at least value for some people who might read the comment. Obviously some people with in-depth knowledge of the paper and field in general will probably already be aware of the link I posted (for example), but others might find it an interesting avenue to explore. I know I often come to HN comments for this reason: a topic seems interesting, I want to learn of related stuff and/or get opinions from people with deeper expertise.
So even if it's only for "future reference", it might actually bring value _for some_. At least that's what I tell myself :)