harshhpareek

Karma: 190

harshhpareek 27 May 2017 7:43 UTC
5 points
on: Looking for machine learning and computer science collaborators
Hi Stuart, I am about to complete a PhD in Machine Learning and would be interested in collaborations like these but only October onwards.

I have written and presented papers at Machine Learning conferences, and am quite interested in contributing to concrete AI safety research. My work so far has been on issues in supervised ranking tasks, but I have read a fair bit on reinforcement learning.

I am not close to Oxford. I am current in Austin, TX and will be in the bay area October onwards.

harshhpareek 16 May 2016 2:57 UTC
12 points
on: Open Thread May 16 - May 22, 2016
Are there any newbies wandering these parts? Leave a comment here! I want to know if there’s any interest in a weekly/fortnightly/monthly newbie thread?

By newbie, I mean that you’ve found LessWrong somewhat recently and are getting exposed to many ideas like those in the Sequences but you don’t post because you wonder if the things you would talk about get discussed later or might have already been discussed somewhere.

LW has a reputation of being very harsh on newbies, so maybe a newbie thread where we can discuss things without annoying those critical people would give people a place to hang out.

harshhpareek 16 May 2016 0:27 UTC
0 points
on: Open Thread May 16 - May 22, 2016
I’ve been having digestive trouble recently and have started wondering if I’ve developed a new allergy/intolerance (Known: milk, cashewnuts, chocolate). Does anyone have a recommendation for tests to check for these?

Apparently, “Eight foods account for 90% of all food-allergic reactions: milk, eggs, peanuts, tree nuts (e.g., walnuts, almonds, cashews, pistachios, pecans), wheat, soy, fish, and shellfish.” (source: http://www.foodallergy.org/file/facts-stats.pdf). However, nuts are good for you (eg. https://examine.com/faq/how-can-i-best-ensure-cardiovascular-health-and-longevity/). So what do you do?

I imagine allergies are bad for your body, even apart from the digestive issues. So, do you take any supplements for nuts? The only one I am aware of here is Omega-3, which is common to various nuts.

Sidenote: If you are willing to put a “nuts without allergies” supplement together, I might buy it from you. See https://www.reddit.com/r/Supplements/comments/3ptsz8/i_am_a_supplement_caffeineenergy_pill_company/cw9gwht for business advice.

harshhpareek 18 Apr 2016 21:08 UTC
4 points
on: Open thread, Apr. 18 - Apr. 24, 2016
I’ve been preparing for coding interviews, and I realized that the skill had gotten “rusty” from disuse. A specific example is coding a binary search, which is a little nontrivial because you have to think carefully to avoid off-by-one errors.

When people talk about old skills they talk about them in two ways: some skills you can supposedly never forget, like riding a bike, Some others can get rusty, so you need to keep brushing them up over and over again.

Neither of these seems actually true. I think it’s more like the exponential forgetting curve we have for (verbal) memory. The neurons for the skill still exist but you can’t access them after a while, and when you recall the skill, you get to the same level as before. If you keep reinforcing it from time to time, say according to the spaced repetition schedule, the skills become permanent. (I’ve made the exponential analogy because it would be cool if motor memory and verbal memory had similar mechanisms, but it’s just a model that I’m familiar with)

Has anyone heard of something like this in the psych literature?

What are your experiences with skills like these that you don’t use as often? Have you made a skill “permanent” through repeated practice.

harshhpareek 27 Mar 2016 2:55 UTC
4 points
on: What can we learn from Microsoft’s Tay, its inflammatory tweets, and its shutdown?
There is an opinion expressed here, that I agree with: http://smerity.com/articles/2016/tayandyou.html TL;dr: No “learning” from interactions on twitter happened. The bot was parroting old training data, because it does not really generate text. The researchers didn’t apply an offensiveness filter at all.

I think this chat bot was performing badly right from the start. It would not make sense to give too much importance to the users it was chatting with, and they did not change its mind. That bit of media sensationalism is BS. Natural language generation is an open problem and almost every method I have seen (not an expert in NLP, but would call myself one in Machine Learning) ends up parroting some of its training text, implying that it is overfitting.

Given this, we should learn nothing about AI from this experiment, only about people’s reaction to it, mainly the media reaction to it. Users’ reaction while talking to AI is well documented.

harshhpareek 27 Mar 2016 2:39 UTC
48 points
in reply to: Huluk’s comment on: Lesswrong 2016 Survey
I took it.

harshhpareek 9 Feb 2016 12:49 UTC
12 points
in reply to: RainbowSpacedancer’s comment on: Open Thread, Feb 8 - Feb 15, 2016
I just attended one too! I am composing a post on this, about halfway done. I’d be interested in a collaboration where we both talk about our experiences, though I would like to see what you think. My post is laden with my own interpretations. Send me a message if you want to discuss once you have your outline down

harshhpareek 8 Feb 2016 16:15 UTC
2 points
in reply to: Gyrodiot’s comment on: Study partner matching thread
Hi, I’m an AI PhD student and I just signed up for the Udacity Deep Learning course. Lets do this!

harshhpareek 4 Feb 2016 15:49 UTC
9 points
on: Upcoming LW Changes
I’m interested in setting up the dev environment. But I’m running into technical issues setting up the VM etc. I expect more such questions will come up. What is the right place to discuss these? Perhaps a channel on the slack? Or do we want something more permanent to help new contributors?

harshhpareek 20 Jan 2015 23:37 UTC
6 points
in reply to: Scott Garrabrant’s comment on: Open thread, Jan. 19 - Jan. 25, 2015
It depends on why I’m making the list.

If I’m making a todo list for a project I’m working on, Workflowy is good because its simple and supports hierarchical lists.

For longer lived stuff where I add and delete stuff like grocery/shopping lists or books to read, I use wunderlist because they have an android app, a standalone windows app and it looks pretty. Browser-based apps annoy me so I like the windows app and the android app is nice to have when I’m actually in the grocery store.

When I’m making a list because I need to be productive and not as a way to plan, I use a paper todolist: http://www.amazon.com/gp/product/B0006HWLW2/ref=oh_aui_detailpage_o08_s00?ie=UTF8&psc=1. Checking things off on paper does wonders for productivity and having the printed thing helps set the mood.

harshhpareek 20 Jan 2015 23:10 UTC
2 points
in reply to: sixes_and_sevens’s comment on: Open thread, Jan. 19 - Jan. 25, 2015
I tried using RSS readers, but I tended to forget to check their websites or apps. I could have trained myself to check them more often but I ended up using https://blogtrottr.com/ instead. It sends RSS feeds to your email inbox, so I can check blogs along with my email in the morning.

I haven’t had any issues so far. They send you ads along with the feed to generate revenue. Having a revenue model is a solid plus in my book.

What I don’t like about it: they don’t have accounts so managing subscriptions is a little hard.

harshhpareek 12 Jan 2015 20:06 UTC
0 points
in reply to: Lumifer’s comment on: Open thread, Jan. 12 - Jan. 18, 2015
I haven’t put sufficient effort into identifying healthy cooking oils. I currently use Crisco’s Blended Oil supplemented with Omega-3. The question is if it is supplemented in the right amount, and that information is not provided.

Animal fats are low in Omega-6 but I think the Omega-3:6 ratio is a problem for meat-eaters too.

harshhpareek 12 Jan 2015 19:43 UTC
9 points
in reply to: sediment’s comment on: Open thread, Jan. 12 - Jan. 18, 2015
I’m a vegetarian and I looked into this stuff a while back. The Examine.com page What beneficial compounds are primarily found in animal products? is a useful reference with sources and includes the ones you wrote above. An older page with some references is this one.

I currently supplement with a multivitamin (this one—Hair, Skin and Nails), creatine and occasionally Coenzyme Q-10 and choline, You didn’t mention the last two but I have subjectively felt they increase alertness. I (hopefully) get my Omega-3/6 fatty acids from cooking oil. I had a basic panel done and found I was deficient in Calcium (probably due to my specific diet, but it is worth mentioning) and B12. So, I supplement for Calcium too.

I do regular exercise (usually bodyweight and dumbbells) and I had disappointing results without whey protein and creatine supplementation. Excessive amounts of creatine (look up “loading”) is recommended for bodybuilders but 5g/day is recommended for vegetarians. See gwern’s review and the examine.com review.. The examine.com review mentions that the fear of this compound is irrational and recommends 5g a day for everyone, pointing out that creatine would have been labeled a vitamin if it wasn’t produced in the body. (Excessive creatine causes stomach upsets but I wasn’t able to find a value at which this happens, and I’ve never experienced this myself).

I also take a fiber supplement, Metamucil. This one isn’t vegetarian-specific, but I highly recommend it.

harshhpareek 9 Dec 2014 6:42 UTC
9 points
on: Open thread, Dec. 8 - Dec. 15, 2014
Are there any LessWrongers at NIPS (in Montreal) this week? Perhaps we can have a mini meetup. Send me a PM or reply if you’re here. I’m here till Sunday.

harshhpareek 19 May 2013 23:38 UTC
1 point
on: Less Wrong Sequences+Website feed app for Android
The link doesn’t work anymore. Can you host it somewhere else?

I did a superficial search and it seems this is the only android app for browsing less wrong. Are there any others?

harshhpareek 4 Mar 2013 2:51 UTC
0 points
in reply to: Eugine_Nier’s comment on: Giving in to small vices
My reply is an answer to “Is there a difference between what’s right and what works?” I’m not trying to open the can of worms labelled “Are draconian deterrents justified?”, merely trying to show through example that what works may fall in moral grey areas. To briefly answer your comments--

If the deterrent works few hands actually need to be cut off

The effectiveness of strong deterrents is questionable#Effectiveness). Considering the links mentioned above, some cases do spring up despite the deterrent, in this particular situation. Speculatively, someone who is unable to make enough money to feed his family otherwise may take this risk.

I think you’re underestimating the benefits of high implicit trust.

There are definitely reasons that the system continues to exist. The “trust” induced by harsh laws is one of them.

harshhpareek 4 Mar 2013 0:35 UTC
0 points
in reply to: DanielLC’s comment on: Giving in to small vices
Consider Saudi Arabia which even today implements the Sharia policy of cutting off the hands of those who steal, especially those who steal during prayer times. I have heard anecdotal stories, that even jewelry shops in Dubai are left unlocked during prayer times. The fear of punishment is so high that no one dares steal.

Does this policy work? Yes. Is it right? Debatable. I would argue that asking people to lock their shops is a smaller cost to society than the cost of fear and of the possible loss of limbs from this procedure, and the benefit—being able to keep shops open—is small. Of course, there is another implicit benefit, that of being consistent with other Sharia values which I think outweighs all the other points here.

harshhpareek 3 Mar 2013 20:05 UTC
32 points
on: Rationality Quotes March 2013

The world of the manager is one of problems and opportunities. Problems are to be managed; one must understand the nature of the problem, amass resources adequate to deal with it, and “work the problem” on an ongoing basis.[...] But what if the problem can be fixed? This is not the domain of the manager.

An engineer believes most problems have solutions. The engineer isn’t interested in building an organisation to cope with the problem. [...] And yet the engineer’s faith in fixes often blinds him to the fact that many problems, especially those involving people, don’t have the kind of complete permanent solutions he seeks.

-- John Walker, The Hacker’s Diet (~loc 250 on an e-reader)

harshhpareek 3 Aug 2012 20:05 UTC
0 points
in reply to: RobertLumley’s comment on: Rationality Quotes August 2012
I meant lukeprog’s karma, i.e. the poster of a comment influences how good the comment is.

harshhpareek 3 Aug 2012 19:49 UTC
0 points
in reply to: JohnEPaton’s comment on: Reinforcement Learning: A Non-Standard Introduction (Part 1)
(Assuming Mind=Brain, i.e. the entire mind is just the physical brain and no “soul” is involved. Also, Neurons aren’t really all that small, they’re quite macroscopic—though the processes in the neurons like chemical interactions need quantum mechanics for their description)

In Newtonian Mechanics, it is sufficient to know the positions and velocities (i.e. derivaties of position) of particles to determine future states. So, the world is Markov given this informatio.

In Schrodinger’s equation, you again only need to know \Psi and it’s time derivative to know all future states. I think the quantum properties of the brain are adequately described just with Schodinger’s equation. You do need to include nuclear forces etc in a description of the brain. You may need quantum electrodynamics, but I think Schrodinger’s equation is sufficient.

My physics education stopped before I got here, but Dirac’s equation which may be necessary to model the brain seems to require the second time-derivative of the wavefunction—so you may need the second order time-derivatives to make the model Markov. Can someone who knows a bit more quantum physics chime in here?

EDIT: Reading the wiki article more carefully, it seems Dirac’s equation is also first order