lazyjeff

This is a listing of best paper awards across 32 computer science conferences, for papers chosen in that year to be the "best" by committees in those conferences meant to be the broadest conferences in each subfield.

I use a single never-ending text file for this, because I'm tired of dealing with remembering formats or apps. I've written about it here: https://jeffhuang.com/productivity_text_file/

I've already made such a tool, and it works for any browser on Windows, and all the data is local -- yours. https://irchiver.com/

So I was thinking about how the web is not as permanent as it used to be. Between walled gardens, and dynamic content, it's hard to find things you saw. So I wanted a way to have a screenshot-based "archive.org" but for yourself, so that it works with non-public content too like a Facebook post or if it disappears like all the stuff on Google+.

Philosophically, I think almost everything we see and think about is in a web browser these days. Even private conversations with my friends, I often look up a few things about them. So I've been using irchiver myself to dogfood it, for almost 2 years now. And found it to be super useful already. For example, I've recovered important technical posts that someone deleted, paragraphs on text I lost when submitting a form that didn't save, or found a comic strip that I had at the tip of my tongue but got lost in google search.

And I think it's now possible because of strong+efficient screenshot compression and good OCR. There's an undocumented Win32 hack that lets you capture window content in a fast way (moreso that a typical BitBlt) so I used that to grab the images. And made a standard inverted index search engine for the content. Anyways, the webpage I put up for the project explains it as best as I can to a general audience [https://irchiver.com/].

In many ways, it's like the Rewind startup here. Except it just happens I did it for Windows, and they did it for MacOS and iPhone. And I care a lot about local storage of the original screenshots and text. Would love to hear others' thoughts.

I've been working on a similar system for a while called irchiver [https://irchiver.com/] but what I wanted from the start is for everything to be local and plain formatted, i.e. stored as .txt and .webp to be most accessible.

Coincidentally, I've only been building for Windows, so different from the platforms supported by Rewind. If anyone wants to collaborate, I'd be very open to it.

There's an open source project I was briefly involved in called SSVG [1] that renders the SVG as Canvas to speed it up drastically, especially on Chrome. It works as a simple one-line js drop in for many common visualization examples [2].

[1] https://ssvg.io/ [2] https://ssvg.io/examples

Before Rewind.ai launched, I had made a counterpart for Windows, but just for web browsers: https://irchiver.com/

Completely local, no network traffic at all.

I've had my ups and downs with peer review. I'm trying to collect stories of papers from ideation to publication, while also describing the struggle for my own papers.

I've written the backstories for 30 of my papers here: https://jeffhuang.com/struggle_for_each_paper/

Looking at my past experiences for all my papers, I can only conclude that the process is more nuanced than can be described in a tagline like "peer review is broken" or "publishing is random", so I'm interested in hearing everyone else's experiences too.

Some of what looks challenging here is handling fonts. SVG would be so much better if there was a way to do font embedding. Even a minimal set of fonts would be amazing. Or if operating systems or libraries for SVG support could all commit to supporting a common set of fonts, like browsers did for web safe fonts. Think of how clunky the web would have been without those.

Are you sure you recall correctly? The linked page recommends upgrading GIF to PNG, "the only reasonable alternative to paying the Unisys tax on the web is to upgrade graphics from GIF to PNG format", which makes sense because it's a lossless conversion. PNG was a new format back that, without much adoption. JPEG doesn't make sense as a substitute format for GIF, and would greatly increase the size and decrease the quality of files.

Spacemonger is probably the oldest piece of non-updated software I still use. I've been using it continuously for over 20 years, about the same length as I've been using Winamp 2, and just recently used it to clear out some large files to create a full-disk backup.

It still works blazing fast on hard drives that are 1000 times the size than the drives that existed when it was programmed, and nearly as well as 20+ years ago (there are some occasional crashes when removing some system protected files, but easy to just open it up and scan again). I never made the jump to Spacemonger 2 because in the early years that I started using it, I was in high school without a credit card.

I emailed the author in December 2020 to see if there was a way to enable HiDPI on it, since that's the only real flaw of it right now. He responded 3 months later with a very personal message that I really appreciated.

I've started doing this more. The videos of my work has always outlasted the work itself. Even though technically I can dig up old compilers or try and update my dependencies, I rarely do. So whatever was captured in the video becomes the only artifact of my older programming projects.

I used to think that video formats would no longer be supported over time, but even the oldest weird video formats still play in VLC and MPC, and probably would work fine if uploaded on YouTube.

If you want automatic archiving, check out https://irchiver.com/ for Windows. It runs on client-side so gets an archive from your point of view, i.e. including Facebook content and partially-filled out forms, etc.

Fair point, it's something I'm aware of and it's a bit intentional to counterbalance the "normalization bias" already done in other rankings, if that's what you meant by "account for differences".

To put another way, to count in a way such that a CVPR best paper is worth 28-fold a CHI best paper is also quite suspect. There's a rabbit hole you can go down to find the best conference to submit to where the submission-to-best-papers is low, to optimize for this.

PS: note that there's an upper bound to best paper awards, which is "< 1% of submitted papers".

I've been looking at the bias in rankings for a little while. I think one way to identify and raise awareness of the biases, is just put rankings together side-by-side. I did this for computer science programs, and there's some interesting differences that I noticed:

https://jeffhuang.com/computer-science-open-data/#bias-in-co...

This is the original analysis by one of Columbia's math professors that led to this investigation: https://www.math.columbia.edu/~thaddeus/ranking/investigatio...

My own reading is that the math professor's analysis is quite credible, and that Columbia did indeed cheat the rankings by reporting "creative statistics".

Nice idea, I could imagine using this. Just a thought, it might be nice to see more variation in the Lorem Ipsum, like with text that include numbers and links and capitalized word, maybe even multiple paragraphs.

irchiver captures text on the page, and separately OCRs the screenshots (specifically, the screenshot from your viewport). So you can search just what was shown on the page, or what was in the page. Both techniques have pros and cons.

While archive.org is fantastic, it can only capture pages that are both 1) publicly accessible (i.e. no social media content) that it happens to crawl, and 2) static content (you're out of luck if the content you want is loaded dynamically, or changes depending on user input).

I've been working on this problem for a while. Website upkeep is hard to quantify, but basically every disk fails and every operating system eventually needs a serious upgrade. The timeframe that a system can run continuously is not that long compared to the timeframe that information is relevant. So the most lightweight way to keep something up and running is to make it trivial to port to many hosting configurations by simplifying the toolchain needed to rehost it. (Note that humans are part of that workflow, if it's a company)

I've written a manifesto about making a commitment to keep websites online and maintained for 10-30 years, for people who are maintaining web content: https://jeffhuang.com/designed_to_last/

And on the flipside (from a user's point of view), I've also been working on a background process that automatically captures full-resolution screenshots of every website you visit, creating your own searchable personal web archive: https://irchiver.com/

I've personally been trying to make a commitment to keep my web projects and writing online for 30 years. My original internal goal when I started thinking about this, was to outlast all the content on Twitter, Google+, and facebook.com. One of those has already been met, kind of sadly.

Because politicians exempted themselved from anti-spam laws, as they do with most laws.

This was the most puzzling thing to me. The politicians that I saw on TV as adamantly pro-privacy, anti-tracking, who made a lot of sense in everything they were saying -- you contribute a single dollar (because they want to show grassroots support for their pro-individuals campaign) and they IMMEDIATELY give your email and survey responses to everyone in their party, including to state-level campaigns in places across the country.

There was no indication on the donation form that any of my personal details would be used for anything except to show that they had a lot of grassroots supporters.

Not only that, but their emails are so clickbait-ey like "lazyjeff, you are the reason that [hated politician] is destroying democracy."

The Leap Motion still does hand tracking better than MediaPipe, and it's still the best hand tracker I know of (besides larger devices like the Oculus Quest).

We've got an open source library for mobile hand-object input [https://portalble.cs.brown.edu/], and the version with Leap Motion is really nice, but doesn't directly work with a phone (we had to pipe data through a compute stick to make it work).

I'd love to see MediaPipe Hands match LeapMotion precision some day, but I'm not even sure if it's possible. A real depth sensor goes a long way.

Thank you for the kind words!

I'm not the developer, but he (Jesse) spoke in my class over a year ago. So hopefully I'm conveying this correctly -- my impression is he felt it was pretty close for live jamming but depends on a lot of conditions (ping latency, ethernet vs wifi, how many people are in the session). But what I can say for sure is it makes conversations feel a lot more intimate.

Thanks for checking about the numbers. I did just double check the Maryland numbers, and what's in my table seems correct:

The 25K you're referring to is for 9.5-months so is normalized to 9 months to compare against the other 9-month numbers. It's already notated with the [1] footnote in the table.

The 36k is not guaranteed in the offers given to students: the UMaryland CS offer letters states "Students who are TAs or RAs over the summer earn an additional $5,600 up to $11,000", which I can only interpret to mean that 11K is not guaranteed. So what's reported in that table is my interpretation of the minimum stipend if the student does get one, but what you say makes sense, that an RA provides the higher number in that range -- I just don't have a way to represent that.

I'm not sure how this would work in practice.

It's already working in practice for almost half the universities: summer stipends are guaranteed, but if students choose to intern elsewhere, you don't get the summer stipend. Nothing changes if their advisor is already funding them. I'm just saying the other half of universities should be doing this as well, that it should be universal.

Yes it does cost a bit, but if you think about what's happening now for universities with no summer guarantee, it's that PhD students without summer internships are getting no pay in the summer. And even the ones that do find summer pay, it's still a stressful situation for them during the academic year, so a complete distraction.

Honestly, I can't think of a better use of university money than to be paying PhD students who would otherwise be working for free in the summer (if their advisor can't fund them).

I recently collected the first-year PhD stipends in computer science across 39 university departments, to have a real sense of the numbers. Basically to have a fair comparison, because a lot of reported stipends aren't clear whether they are 9-month or 12-month stipends, and stipends can different depending on the year/candidacy of the student.

Data: https://jeffhuang.com/computer-science-open-data/#verified-c...

The main conclusions I arrived at from this exercise is that 1) CS stipends vary quite a bit but have slowly been increasing, but NSF funding needs to catch up for them to go higher, 2) the 12-month stipend makes a big difference from the 9-month stipend, and I think universities should guarantee the 12-month stipend (which only half do), rather than making summer stipend dependent on availability of advisor funding or student jobs (like TAing).

My phone already has av1 hardware decoding (pixel 6). But anyway it's a chicken/egg problem, where the broader hardware decoding will come once browser support is there. The long-term impacts are more significant.

Also, energy use worldwide might decline if fewer datacenters are needed to store/transmit videos. Kind of hard to measure and compare against enduser devices, but it's another factor to consider.

While we're making wishlists for Safari, the one I think would have the most real world impact is AV1 support, to have a modern royalty-free codec across all browsers if Safari adds it. Safari is the last holdout. Considering the large proportion of internet usage is to transfer video, the savings in storage and transmission would be non-trivial.

It's definitely a consideration. The way I think of AR is:

AR/VR headset = desktop

AR glasses (2025+) = laptop

phone AR = phone

So phone AR has ergonomic challenges kind of like the small screen and slow speeds that smartphones had when they first came out. You'd use phone AR to throw a pokeball outside, or use phone AR to touch up your virtual makeup / decor for AR selfies like you do with phone selfies (which are similarly ergonomically challenging). When AR glasses become an everyday thing, they'll be the workhorses and you'll take it around with you while traveling or for work. Finally, for dedicated production tasks, you'll get your AR/VR headsets and use it in pre-specified locations.

So I'm a minor contributor to the posted project. And my personal opinion is that phone AR and headset AR will be completely different use cases.

Phones are everywhere because they fit in your pocket. Headsets never will, so will be specialty hardware. Maybe I'm wrong and it'll be normal to walk around with headsets everywhere, but I just can't see it happening.

Phone AR is technologically worse, and the version we released today is performance-wise not as good as the version with extra hardware we released 3 years ago. But it's slowly improving, and convenient form factors often beat more cumbersome devices over the long term, e.g. phones vs desktops. So I they can exist side-by-side, but headsets will be only used in some scenarios, while phones will be used by everyone.