All of VipulNaik's Comments + Replies
One particularly amusing bug I was involved with was with an early version of the content recommendation engine at the company I worked at (this is used by websites to recommend related content on the website, such as related videos, articles, etc.). One of the customers for the recommendation engine was a music video service, and we/they noticed that One Direction's song called Infinity was showing up at the top of our recommendations a little too often. (I think this was triggered by the release of another One Direction song bringing the Infinity song in...
He points to recent events at Sports Illustrated. But to me the SI incident was the opposite. It indicated that we cannot do this yet. The AI content is not good. Not yet. Nor are we especially close. Instead people are using AI to produce garbage that fools us into clicking on it. How close are we to the AI content actually being as good as the human content? Good question.
I happen to work at the company (The Arena Group) that operates Sports Illustrated. I wasn't involved with any of this stuff directly (and may not even have been at the company ov...
Thanks for reviewing and catching these subtle issues!
Technically, it is not true that the prime numbers being multiplied need to be distinct. For example, 2*2=4 is the product of two prime numbers, but it is not the product of two distinct prime numbers.
Good point, I've marked this as an error. My prompt about gcd did specify distinctness but the prompt about product did not, so this is indeed an error.
This seems wrong: "neither can be definitively identified" makes it sound like they exist but just can't be identified...
I passed on this one as be...
Thanks, fixed now! Sorry I missed that.
Good question!
First, my original post didn't provide the correct answers for most questions, only what was wrong with ChatGPT. Going from knowing what was wrong to actually giving correct answers seems like quite a leap. Further, ChatGPT changed its answers to better ones (including more rigorous explanations) even in cases where its original answers were correct.
Second, ChatGPT's self-reported training data cutoff date at the time it was asked these questions was September 2021 or January 2022. To my knowledge, Issa didn't ask it this question, but source...
Thanks, and sorry I missed that error. I've updated the post by bolding the error, and also HT'ed your contribution.
Regarding the topic of batteries getting better while also being harder to remove/replace, I found these interesting comments by FeRD (https://www.ifixit.com/Wiki/What_to_do_with_a_swollen_battery?permalink=comment-911590#comment-911590 + the next two comments) that I quote in full below:
...(1/2) I doubt it's entirely fair to blame Apple for this; they may not have even been the first manufacturer to use a non-removable battery pack. They were certainly the first biggest company to make that switch, but plenty of others did the same, and in some cases way t
A few thoughts:
- In at least one area, namely cars, durability relative to actual usage has improved a lot over the past 50 years or so. See for instance https://en.wikipedia.org/wiki/Car_longevity "According to the New York Times, in the 1960s and 1970s, the typical car reached its end of life around 100,000 miles (160,000 km), but due to manufacturing improvements in the 2000s, such as tighter tolerances and better anti-corrosion coatings, the typical car lasts closer to 200,000 miles (320,000 km)."
- The area where I'm most aware of claims of reduced dura
1Hmm, aren't exposure notifications an opt-in program? I was never forced to get them -- I chose to download and install the app and keep it on. The same way I choose to allow Google Maps to keep a record of my physical location.
I added logs of two further ChatGPT sessions, one of which repeated many of the prompts I used here, tested against the 2023-01-09 version of ChatGPT: https://github.com/vipulnaik/working-drafts/commit/427f5997d48d78c69e3e16eeca99f0b22dc3ffd3
I had originally been thinking of formatting these into a blog post or posts, and I might still do so, but probably not for the next two months, so just sharing the raw logs for now so that people reading the comments on this post see my update.
Good point -- I removed the tag!
Somewhat related, though different in various ways, is this post by Bryan Caplan: https://www.econlib.org/the-cause-of-what-i-feel-is-what-i-do-how-i-eliminate-pain/
I'm curious to hear examples of other worthwhile things in the direction that you have in mind!
Thanks -- good points and well-presented with precision and flair!
Good point! It could be that both kinds of mental exercise (excess stimulation and lack of stimulation) are important for building mental strength; modern society provides the former in abundance (and particularly so for LessWrong readers!), so the form of exercise we're constrained on is the lack-of-stimulation kind (and that's where meditation helps). How far-fetched does that sound?
An anonymous friend to whom I sent this post writes:
...He has a good point that most people just want to do the universal “safety” precautions. I think a big reason that he doesn’t mention is that reasonable precautions are how all businesses defend themselves from lawsuits (e.g. sexual harassment and DEI training); as long as they take the reasonable precautions, then they are immune from lawsuits. But I don’t buy “safety” as an explanation for what policies are possible. It sounds like a just-so story for why we are in the mess that we are in. Vaccines ar
I just got my booster dose today (December 24) and intend to monitor closely. I'll be regularly updating https://github.com/vipulnaik/diet-exercise-health/blob/master/notes/2021-12-24-pfizer-covid-vaccine-booster-dose.md with temperature readings and subjective details of my experience.
I did similar logging after the second dose, that you can see here: https://github.com/vipulnaik/diet-exercise-health/blob/master/notes/2021-06-25-pfizer-covid-vaccine-dose-2.md
By "this way" do you mean the way I wrote it or the way Alexei would have preferred?
Thanks for the feedback! It seems like you're saying I should first have done "negotiation techniques" then "do these negotiation techniques have a place in rational discourse?" as separate sections. So if we make a table with rows as techniques and columns as lenses, then I should have traversed it column major instead of row major.
Did I misunderstand or miss an angle to what you're saying?
Good point! Voss talks a bit about how many of these techniques feel odd. Two points he makes:
- Practice in low-stakes situations to get more comfortable with it. Don't try any negotiation technique in a high-stakes situation that you don't have practice with!
- In many cases the discomfort you experience saying it isn't noticed by others. Voss gives examples related to mirroring as well as to the calibrated question "How am I supposed to do that?" People feel apprehensive asking the question but it usually works despite their apprehension.
I would also add...
We cover a larger period in the overall summary and full timeline. The summary by year starts 2013 because (it appears that) that's around the time that enough started happening per year. Though we might expand it a little further to the past as we continue to expand the timeline.
<describe lockdowns as social engineering>
Did you intend to expand this?
<Michael Mina stuff here>
Did you intend to expand this?
Do you mean PDT instead of PST in the title?
I did some rewording of the post that made it a little more wordy, but fingers crossed that that part has now become less confusing.
Thank you for the feedback (and also for discussing this at length which gave me better understanding of the nuances). I modified to a more clumsy but hopefully a more what-you-see-is-what-I-mean term: https://www.lesswrong.com/posts/mRkWTpH9mb8Wdpcn5/coronavirus-california-case-growth?commentId=GHSEwZwR2TSkyzpdm
Thank you for the feedback. I agree with Lukas Gloor's reply below that the choice of term is confusing as it differs from what people may intuitively think "true cases" means. I also agree with his remark that setting terminology that is consistent with reality isn't bad in and of itself.
I have therefore changed "true cases" to "true currently-or-eventually-symptomatic cases". I think that provides the level of precision needed for our purposes. I haven't found a better term after some searching (though not a lot); however, I'm happy to change to a more concise and medically accepted term if I get to learn of one.
What I wrote there was assuming that the number of new true cases drops to a fairly low level. Whether that happens now or a week or two or three later is unclear; if the 2 -> 3 backlog is growing. then resolving that backlog will add more delay.
I posited us already being at this point as the "optimistic" scenario.
I'll reword the post to clarify this.
Directly visiting http://foretold.io gives an ERR_NAME_NOT_RESOLVED. Can you make it so that foretold.io redirects to www.foretold.io?
That's a normal part of life :). Any things that I decide to do in a future day, I'll copy/paste to over there, but I usually won't delete the items from the checklist for the day where I didn't complete them (thereby creating a record of things I expected or hoped to do, but didn't).
For instance, at https://github.com/vipulnaik/daily-updates/issues/54 I have two undone items.
There is some related stuff by Carl Shulman here: https://www.greaterwrong.com/posts/QSHwKqyY4GAXKi9tX/a-personal-history-of-involvement-with-effective-altruism#comment-h9YpvcjaLxpr4hd22 that largely agrees with what I said.
My understanding is that Against Malaria Foundation is a relatively small player in the space of ending malaria, and it's not clear the funders who wish to make a significant dent in malaria would choose to donate to AMF.
One of the reasons GiveWell chose AMF is that there's a clear marginal value of small donation amounts in AMF's operational model -- with a few extra million dollars they can finance bednet distribution in another region. It's not necessarily that AMF itself is the most effective charity to donate to to end malaria -- i...
I tried looking in the IRS Form 990 dataset on Amazon S3, specifically searching the text files for forms published in 2017 and 2016.
I found no match for (case-insensitive) openai (other than one organization that was clearly different, its name had openair in it). Searching (case-insensitive) "open ai" gave matches that all had "open air" or "open aid" in them. So, it seems like either they have a really weird legal name or their Form 990 has not yet been released. Googling didn't reveal any articles of incorporation or legal...
In my experience, writing full-fledged, thoroughly researched material is pretty time-consuming, and if you push that out to the audience immediately, (1) you've sunk a lot of time and effort that the audience may not appreciate or care about, and (2) you might have too large an inferential gap with the audience for them to meaningfully engage.
The alternative I've been toying with is something like this: when I'm roughly halfway through an investigation, I publish a short post that describes my tentative conclusions, without fully rigorous b
FWIW, my impression is that data on Wikipedia has gotten somewhat more accurate over time, due to the push for more citations, though I think much of this effect occurred before the decline started. I think the push for accuracy has traded off a lot against growth of content (both growth in number of pages and growth in amount of data on each page). These are crude impressions (I've read some relevant research but don't have strong reason to believe that should be decisive in this evaluation) but I'm curious to hear what specific impressions you have that are contrary to this.
If you have more fine-grained data at your disposal on different topics and how much each has grown or shrunk in terms of number of pages, data available on each page, and accuracy, please share :).
In the case of Wikipedia, I think the aspects of quality that correlate most with explaining pageviews are readily proxied by quantity. Specifically, the main quality factors in people reading a Wikipedia page are (a) the existence of the page (!), (b) whether the page has the stuff they were looking for. I proxied the first by number of pages, and the second by length of the pages that already existed. Admittedly, there are a lot more subtleties to quality measurement (which I can go into in depth at some other point) some of which can have indirect, long
Great point. As somebody who has been in the crosshairs of Wikipedia mods (see ANI) my bias would push me to agree :). However, despite what I see as problems with Wikipedia mod culture, it remains true that Wikipedia has grown quite a bit, both in number of articles and length of already existing articles, over the time period when pageviews declined. I suspect the culture is probably a factor in that it represents an opportunity cost: a better culture might have led to an (even) better Wikipedia that would not have declined in pageviews so much, but I do
Great points. As I noted in the post, search and social media are the two most likely proximal mechanisms of causation for the part of the decline that's real. But neither may represent the "ultimate" cause: the growth of alternate content sources, or better marketing by them, or changes in user habits, might be what's driving the changes in social media and search traffic patterns (in the sense that the reason Google's showing different results, or Facebook is making some content easier to share, is itself driven by some combinati
The Wikimedia Foundation has not ignored the decline. For instance, they discuss the overall trends in detail in their quarterly readership metrics reports, the latest of which is at https://commons.wikimedia.org/wiki/File:Wikimedia_Foundation_Readers_metrics_Q4_2016-17_(Apr-Jun_2017).pdf The main difference between what they cover and what I intend to cover are (a) they only cover overall rather than per-page pageviews, (b) they focus more on year-over-year comparisons than long-run trends, (c) related to (b), they don't discuss the long-run causes.
They still show up in the total comment count :).
For all the talk about the "decline" of LessWrong, total pageviews and sessions to LessWrong have stayed 5-10 times higher than those to the Effective Altruism Forum (the EAF numbers are documented in my post).
The 2017 SSC Survey had 5500 respondents. Presumably this survey was more widely visible and available than mine (which was one link in the middle of a long link list).
https://slatestarcodex.com/2017/03/17/ssc-survey-2017-results/
Varies heavily by context. Typical alternatives:
(a) Google's own answers for simple questions.
(b) Transactional websites for search terms that denote possible purchase intent, or other websites that are action-oriented (e.g., Yelp reviews).
(c) More "user-friendly" explanation sites (e.g., for medical terminology, a website that explains it in a more friendly style, or WikiHow)
(d) Subject-specific references (some overlap with (c), but could also include domain Wikias, or other wikis)
(e) When the search term is trending because of a recent news item, then links to the news item (even if the search query itself does not specify the associated news)
Interesting. I suspect that even among verbal elites, there are further splits in the type of consumption. Some people are heavy on reading books since they want a full, cohesive story of what's happening, whereas others consume information in smaller bits, building pieces of knowledge across different domains. The latter would probably use Wikipedia more.
Similarly, some people like opinion-rich material whereas others want factual summaries more. The factual summary camp probably uses Wikipedia more.
However, I don't know if there are easy ways of segmenti...
Good idea, but I don't think he does the census that frequently. The most recent one I can find is from 2014: http://slatestarcodex.com/2015/11/04/2014-ssc-survey-results/
The annual LessWrong survey might be another place to consider putting it. I don't know who's responsible for doing it in 2017, but when I find out I'll ask them.
It's not too late, if I do so decide :). In other words, it's always possible to spend later for larger samples, if that actually turns out to be something I want to do.
Right now, I think that:
It'll be pretty expensive: I'd probably want to spend using several different survey tools, since each has its strengths and weaknesses (so SurveyMonkey, Google Surveys, maybe Survata and Mechanical Turk as well). Then with each I'd need 1000+ responses to be able to regress against all variables and variable pairs. The costs do add up quickly to over a thousand do
The first 10 minutes or so of https://podcasts.apple.com/us/podcast/episode-51-what-happened-at-the-arena-group-former/id1615989259?i=1000639450752 also provide related context.