zxexz

16 ms·

1. 211.▲
   by zxexz 2y ago

   I use OpenMPI with no issues over multiple H100 nodes and A100 nodes, with multiple infiniband 200G and ethernet 100G/200G networks, and RDMA (though using mellanox instead of broadcom cards, but afaik broadcom supports this just the s
2. 212.▲
   by zxexz 2y ago

   "Jury is still" out might be a bit strong. I can come up with dozens of random postulates, but I can't magically conjure up a jury to hear them out.
3. 213.▲
   by zxexz 2y ago

   Yep, and "Even women [...]", as if talking about a a curious subset of an eclectic animal species. Beyond that, the transactional wording! Alien archaeologists have some serious linguistic challenges up their sleeve(s). Sorry you&
4. 214.▲
   by zxexz 2y ago

   Y'know, if worded differently, without misogynistic overtones, this could have been a good discussion. Please read the guidelines [0]. [0] https://news.ycombinator.com/newsguidelines.html
5. 215.▲
   by zxexz 2y ago

   Ideally if you do this, you store the key in a separate schema with proper roles so that you can call encrypt() with the database role, which can't select the key. Even then, the decrypted metadata should not be particularly sensitiv
6. 216.▲
   by zxexz 2y ago

   I wonder if PA, OH, and other states with large Amish and Mennonite populations do something similar. Seems a thoughtful idea.
7. 217.▲
   by zxexz 2y ago

   Huh. I feel like that almost daily nowadays. Maybe I should grab some tinfoil. Unless it's coming from inside the head...
8. 218.▲
   by zxexz 2y ago

   Is this an ad for a new, closed-source, GPGPU backend?
9. 219.▲
   by zxexz 2y ago

   Why Ubuntu specifically? Not even being snarky. Calling out a specific distro, vs. the operating system itself. I've had more pain setting up ML environments with Ubuntu than a MacBook, personally - though pure Debian has been the easi
10. 220.▲
    by zxexz 2y ago

    I was more put off more by the initial example. Amber being used to calculate an age being less than 18. With sed. Seems almost like an attempt at humor, except the bit where one would laugh.
11. 221.▲
    by zxexz 2y ago

    Also, the plot of the wonderful 2008 rock opera, “Repo! The Genetic Opera”. https://en.wikipedia.org/wiki/Repo!_The_Genetic_Opera
12. 222.▲
    by zxexz 2y ago

    I mean, you are using a service they're providing - many would say they they're exercising their rights by gatekeeping how it's used. There are pretty good models out there you could use however you want for your own purpose,
13. 223.▲
    by zxexz 2y ago

    This is quite nice, but you really ought to have some page accessible with attributions to the open source projects you're using to power this!
14. 224.▲
    by zxexz 3y ago

    Or just set up your own recorder at each sign and replay the last week’s traffic in random order.
15. 225.▲
    by zxexz 3y ago

    I wouldn't assume that's intentional on the poster's part. HN should probably just add a rule for Colab notebooks as another top-level comment to this thread pointed out.
16. 226.▲
    by zxexz 3y ago

    It's pretty common for someone to read an article and find it interesting, c/p the title, and submit it with a link, so I assume the former.
17. 227.▲
    by zxexz 3y ago

    Seems pretty reasonable! Not sure that's the easiest path for them to maximize profit though. I'd probably say they should: - allocate the $500M to the new chip, $100M to each of AMD and NVIDIA, then: - never officially hire any
18. 228.▲
    by zxexz 3y ago

    It would be really cool if this became a trend. I'm sure tons of people have printouts from back then.
19. 229.▲
    by zxexz 3y ago

    Assuming US, based on terminology used. Have you considered teaching at community college? Depending on the state, the degree requirement may be dropped altogether, and I'd argue you could "give back" just as much, if not mor
20. 230.▲
    by zxexz 3y ago

    I saw that. The WandB report is here [0]. The losses are quite close, but in my mind we'd need to see parameter counts at least 1 and 2 params more to make conclusions (with the dataset scaling proportionally). If the training performa
21. 231.▲
    GateLoop: Data-Controlled Linear Recurrence for Sequence Modeling(arxiv.org)

    26 points by zxexz 3y ago|2 comments
22. 232.▲
    by zxexz 3y ago

    The simulations aren't nested, they're simultaneous and all under one hypervisor! Maybe one of my old Qubes laptops is running the second layer of the simulation - I should probably turn it back on.
23. 233.▲
    by zxexz 3y ago

    I don't doubt he's serious in what he believes. I respect Ilya greatly as a researcher. Do you have notes or a time-point for me to listen to in that podcast? I bemoan the trend toward podcasts-as-references - even a time-point re
24. 234.▲
    by zxexz 3y ago

    Pretty sure this is tongue-in-cheek, given the commenter (pretty high profile dev who creates a lot of cool stuff and does not shy away from using other open source code!) :)
25. 235.▲
    by zxexz 3y ago

    For what it's worth, the MIT Technology Review these days is considered to be closer to a "tech tabloid" than an actual news source. I personally would find it hard to believe (on gut instinct, nothing empirical) that AGI has
26. 236.▲
    by zxexz 3y ago

    I'm not saying that he hasn't. I'm saying the linked tweet does not offer evidence.
27. 237.▲
    by zxexz 3y ago

    There has been no confirmation of him returning. This is an emoji response to a tweet that does not convey any information.
28. 238.▲
    by zxexz 3y ago

    Maybe yes, maybe no...you're reposting a tweet that does not say that at all.
29. 239.▲
    by zxexz 3y ago

    I just want at least 1 of both. For at least the next 10 years. I will still have USB-A peripherals I use daily to weekly until then. I would've paid $100+ extra for a USB-A port on this macbook.
30. 240.▲
    by zxexz 3y ago

    Hey man - not trying to get on the bad side of one of my favourite databases! Making an observation, one that others have made as well.