ghgr

Interesting concept. So you propose to have a landing page with a selection of videos tailored to the taste of the visitor? And seamless combining results from YouTube, Vimeo and others?

You would be effectively relegating YouTube to a commoditized CDN for videos.

Again, interesting concept.

Exactly. I'm constantly surprised how many companies carelessly upload all their internal documents to Google Docs, S3, Dropbox, GitHub, Slack et al. I'd suggest that anything that's not supposed to be public never be uploaded to such services, specially if you are not from the US. The way to go even for small tech companies is self hosting, except maybe for email, due to GMail marking your emails as spam from what I've heard.

If you are based in Europe drop me an email. We might have something for you.

I'm sure GPT-3 is impressive and probably a paradigm change, but all this hype has an air of urban legend that somehow irks me. We all know of somebody's friend who had access and blew its mind, and that is "too dangerous" to be released to the world. There's an API, but not public, and the only way to get access is to join a waitlist (and wait).

If it were a marketing ploy it would be indeed terrific. It reminds me of the movie "The Blair Witch Project". And I can't wait to try it by myself.

> Unsavory Use-cases

> Tunshell could also be used as an exploitation tool to gain unauthorized access to remote hosts. Personally, I hope that this tool is not misused for nefarious purposes. If it becomes apparent that tunshell is helping malicious actors go about their activities, the free service will be discontinued.

It seems the author already though about that. Well, don't we all here have the Security Mindset (TM)? [1]

[1] https://www.schneier.com/blog/archives/2008/03/the_security_...

The other side of the coin are the inflated traffic statistics that include all kind of bots & crawlers with spoofed user-agent, and for low traffic sites like niche personal blogs with a custom domain it can be 90+% of the server-side logged visits.

How proud was the 15-year old me with my first .com domain, having over 100 visitors per day. Little did I know that the actual number of visitors was much, much less than that.

Wow, the one about the DNA is almost creepy how accurate it is. It even knows what an "ancestry site" is! Not to mention how consistent the story is.

It seems to be down (HTTP/2 503)

By the way, has somebody already tried to plug the posts from /r/WritingPrompts/ into GPT-3 and check the results?

Thanks for sharing. Since it's your own repository I'd recommend to append a "Show HN:" at the beginning.

I've also learned of Netron to visualize a network's architecture. That's neat way to get a grasp of a new model (here a screenshot: https://raw.githubusercontent.com/lutzroeder/netron/master/....)

I've been wondering if using such exhaustive datasets would allow to discover the existing supply chain routes (which is usually a closed guarded trade secret). For that I would model the truck, ship and freight train traffic as edges, and mines/factories/companies as nodes in a graph. The weight of each edge is proportional to the number of containers that flow through it per unit of time. The question is, would it be possible to statistically infer the interdependence among nodes?

I started it and so far I got nice visualizations of air and maritime traffic. I should resume it when I find some time.

At my company we use a self-hosted Nextcloud instance with the Talk plug-in. It has chat, video, audio and screen-sharing. It may not be E2E encrypted, but the data never leaves our servers. It also makes a better impression when you share a <own_company>.eu link instead of a .webex.com or .zoom.us link.

Direct link (since it's not easy to find in the article):

https://e.foundation/

I mostly agree with you that there are many false positives in the research papers. Still, you shouldn't outright dismiss the possibility of you not implementing their models correctly.

> How do you plan to prevent abuse and fraud? This seems like it could attract a great amount of misuse.

Exactly. In addition, in some countries (e.g. Germany) your client is required to have a physical address [1] not only in the country, but also in the area where they get their number.

[1] "Address" being a place where they can receive snail mail AFAIK

That's the problem I have when making a toy project from scratch in order to learn a technique (be it databases, blockchain, or deep learning). I'm always wondering if I'm allowed to use such algorithm or clone such github repo. Or if copy/pasting an auxiliary function is cheating or not.

To avoid such moral dilemmas I try to come up with projects that solve a new problem (which also comes with its own set of challenges). And then use all the means at my disposal.

As a bonus it also keeps me motivated, since it has a shot at being a profitable business idea.

I didn't compress an actual image from 216GB to 17GB. Instead, I created the highest resolution image already as a set of tiles (each one 5000x10000 px). Then I tiled them further in two additional zoom levels, until I got 4x8 square tiles of 1250x1250px. From there I worked backwards merging chunks of 2x2 tiles.

The figure 217 GB comes from the cumulative size of all the pandas DataFrames containing the whole dataset (as mgspack files). The lossless compressed figure comes from the total size of the highest zoom level.

It didn't take that long, it was mostly done overnight (so around 10 hours in a i7 server with 24 GB RAM).

I followed a tiled based approach, as explained here: https://openlayers.org/en/latest/examples/localized-openstre...

Although the image is not a map at all, I tiled it and served it as a custom map layer. You can view it at https://finance.alisio-computing.eu

What a coincidence! I discovered OpenLayers yesterday when looking for a way to display a 1 Terapixel image. I got it up and running in a few hours. Kudos to the developers!

Previous submission/discussion (without paywall): https://www.piper.com/press-releases/piper-announces-new-m60...

Indeed GANs are showing very promising results. There's a series of blog posts from Fernando de Meer which discuss this topic in a very approachable way. https://quantdare.com/generating-financial-series-with-gener...

CEO & Founder here:

After working for 5+ years at the German research institute Fraunhofer I've noticed the gap between research and business applications. So I founded Alisio Computing to provide businesses with a hand curated choice of state-of-the-art algorithms. From the automotive industry I learned how important is for B2B the seriousness and dependability, so that's why I offer as unique selling point an EU based company happily following the draconian data protection rules of the GDPR. Also, although registered as a German GmbH, all European companies are first class clients and we'll handle your requests in English, German, French and Spanish.

And after this elevator speech, I introduce myself. My name is Eduardo Pena Vina, long time member of HN and even longer time lurker. With Alisio Computing I built the service I'd loved it existed when I was at Fraunhofer. So tired of waiting for somebody else to do it, I built it myself.

It was a challenge to build a dependency based distributed scheduler for batch inference tasks. I'm excited to see how it behaves under real load. :) I welcome all your feedback.

Wouldn't a solution be to install Lineage OS [1] or similar custom ROM and have the advantages of a cheap phone and the privacy of a high-end one?

[1] https://wiki.lineageos.org/devices/

I see it only works on Windows :( Are you planning on compiling it for other systems, or even as a webapp?

By the way, does the computer running it requires CUDA?

> the very term A/B testing refers to the "A" and "B" sides of a newspaper print roller being able to print two versions of an advertisement.

Really? That's interesting. But how could they tell which version converted better? Maybe distributing each version in a different area?

For those of you interested in bioinformatics I can recommend the Rosalind Project [1]. It's like the Euler Project, but for Bioinformatics.

[1] http://rosalind.info/problems/locations/

Restated once more: "Would you trade free time in your youth for the same amount of free time at retirement age?"

Nice project! In a similar vein there is this project [1] who claims online end to end learning in a Raspberry Pi (included training).

[1] https://youtu.be/0ibVhtuQkZA

OP here. My question was not so much about finding collaborators for my specific projects, but rather to find a website where people shared theirs. The examples I gave are just the kind of projects I'd like to find in such a site.

The points you raised (what to get out of the project, communication methods, time investment...) is the kind of information that would fit in the "Description" section. Ideally (for me) would be a site you browse, find a project that catch your eye (big data? deep learning? bioinformatics? etc.) and you jump in, contributing what you believe you can (and learning in the way). No long term plan, just plain old hacking.

There's been work (in a related field) to "stabilize" the jitter in consecutive frames. As you say, by taking into account the neighboring frames.

Relevant excerpt from [1]:

“If you just apply the algorithm frame by frame, you don’t get a coherent video — you get flickering in the sequence,” says University of Freiburg postdoc Alexey Dosovitskiy. “What we do is introduce additional constraints, which make the video consistent.”

[1] https://blogs.nvidia.com/blog/2016/05/25/deep-learning-paint...

Interesting how the New York Times is following the Taboola-like structure of clickbaity headlines. Maybe it's where the money is after all.