Hacker Newsnew | past | comments | ask | show | jobs | submit | more pixelHD's commentslogin

I had my own mail server running for close to a year after which i gave up because i started seeing issues delivering mail to gmail and apple mail accounts. I had to keep tinkering with the DKIM/DMARC/SPF entries to make sure i'm compliant with gmail/outlook/etc. At somepoint i gave up because it wasn't worth the investment of time.

FWIW, i used mail cow [0], which was a delight to setup and use. And it comes bundled with a decent-ish material design webclient.

I would love to go back to hosting my own mail server though. As of now, i shut it down, and use gsuite.

[0]: https://mailcow.email/


Still running my mailcow. I don't care if people don't get my mail. I can show them that i handed the over to their server - everything else is their concern. At least legally. If you can't read my mail cause you misconfigured your spam filter I can't help.


> I don't care if people don't get my mail

This is the type of nonsense I expect from HN. Well done.


When someone in a company agrees on a meeting in their office with someone from outside the company, and the visitor arrives at time well-dressed and is rejected by the companies security without any explanation, then who is at fault? It is not the visitor, it is the awful cooperation between the person in the company and their companies security.

It is entirely reasonable to show how you've handed your mail to gmail and they refused it without actionable explanation, so it is a problem between the recipient and gmail.


GMail being at fault isn't going to make up for a missed invitation (or RSVP) to a special event, or not getting a note of congratulations on an accomplishment, or not getting a message of condolences.

Assigning blame is done after the fact to make up for a mistake.

Just avoid the mistake.


If one completely disregards political consequences, and generally anything but personal short-term benefit, I agree with your reasoning. Otherwise, I see not how that would be possible. And I think arguing just for short-term practicality is deeply flawed; by that you'd also cooperate with a murdering dictator if it helped you personally.


You can run into this on any service. Colleagues on Office365 got mail filtered by spam. There are false positives on every service.

If I get an explanation from the server I will fix it. I get dmarc reports and I make sure all pass. That's all I can do.


I still receive my own email but for sending I use a service or my isp. Unfortunately gmail only supports bigger players unless you get lucky.


I made a discord chatbot for interacting with gpt 2: https://github.com/itsmehemant123/gpt2-discord-bot


I use hopper [0] for something similar. But considering google's scale and resources, google might give me more accurate predictions.

[0]: https://www.hopper.com/


I'd be hesitant on that one, or any prediction software, without the backing of the price guarantee like Google has. Otherwise you're really just gambling.

I used to use whatever flight prediction software Bing bought out when I traveled for work. It was pretty accurate most of the time, and I got used to trusting it, but I definitely got burned really badly a few times by continuing to wait as the price sky rocketed under the algorithms assurance that it would drop again before I needed to pull the trigger.


Pretty sure it was called farecast.com


Isn’t it the opposite - time slows down for you when you move at relativistic speeds, and hence you end up getting to alpha Centauri in 5years w.r.t clocks at Centauri or earth?


An object's relative speed to itself is always 0 (by definition) and so the passage of time is always constant for every object. As you can guess by the name of the theory there is also no "universal clock" or "0 speed" in the universe to compare against, it's only possible to measure speed and the flow of time relatively between objects.

For these reasons "time slows down for you when you move at relativistic speeds" is really a bad way to think of it (and not how the math of the theory actually works). It's much more proper to say things in the form "To <observer a> it appeared as if the clock of <observer b> was <sped up|slowed down>". E.g.:

To the traveler it appeared as if the clock of Alpha Centauri was sped up. To Alpha Centauri it appeared as if the clock of the traveler was slowed down.


It's pretty confusing but essentially, as you travel faster, your clock slows down arbitrarily. In the most extreme case, at the speed of light, your clock stops. For a photon, it emerges and appears at its final location instantly, even if those distances are millions of lightyears apart.

So if you travel to Centauri from Earth, the fastest you get can get there is 5 years as seen from Earth but to the actual spaceship, it could perceive any arbitrarily small change in time.


Oh wow, the newer model is so much better!

If anyone is interested, I have a repo for using it as a discord bot [0].

I didn't know that openAI released a newer larger version though, good to know!

[0]: https://github.com/itsmehemant123/gpt2-discord-bot


I use it in our lab, and it works very well. However, there's no multi-user support. So anyone who connects to the server gets served the editor with the same workspace.

That being said, I think they have the multi-user support in the roadmap.


Interesting. You might want to try gitpod.io to have multi-user support.


Now that nvidia has released a cheaper jetson board, I'm really itching to get it and build a small RL car. Training on the device would be out of the question though.

Anyone know of any successful DIY small RL cars?


The transformer paper was quite influential in machine translation space. This resource [0] posted here a while back is a good place to learn and get a better idea how it works.

[0]: http://nlp.seas.harvard.edu/2018/04/03/attention.html


one of the best visual tutorials on the transformer I came across http://jalammar.github.io/illustrated-transformer/


Wow, that does look really good! Thanks!


you're welcome! :)


Another perspective in similar veins would be the rise of AutoML. Given its absurdly high computational cost, I'd think only enterprises with massive computational power at their disposal would be able to use it.


I've had to use both pytorch and tensorflow for various purposes, and i felt pytorch was more pythonic. In the end, you can use either to implement an architecture, but I liked pytorch quite a bit.


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: