HN user

jaster

361 karma
Posts0
Comments64
View on HN
No posts found.

All the "Thanks for watching!" gave me a good chuckle.

Remind me of one of my own experiences with one of the Whisper model, where some random noise in the middle of the conversation was translated into "Don't forget to like and subscribe".

Really illustrate where the training data is coming from.

Here is a relevant tidbit from the RWKV paper "Limitations" section (https://arxiv.org/abs/2305.13048):

  First, the linear attention of RWKV leads to significant efficiency gains but still, it may also limit the model’s performance on tasks that require recalling minutiae information over very long contexts. This is due to the funneling of information through a single vector representation over many time steps, compared with the full information maintained by the quadratic attention of standard Transformers. In other words, the model’s recurrent architecture inherently limits its ability to “look back” at previous tokens, as opposed to traditional self-attention mechanisms. While learned time decay helps prevent the loss of information,it is mechanistically limited compared to full self-attention.

For me it was worked relatively well.

I track my config in a git repo, and doom-emacs is rather helpful on what external tools must be installed for various configuration "layers" (like when working with such and such language). When using LSP, it can even proposes to install an appropriate language server in some cases.

Sorry, to clarify: I used the package `emacs-next` from guix to install emacs 28. You still need to install and configure emacs-lsp along with the LSP backend(s) you want to use.

I'm currently using the doom-emacs "emacs distro" (https://github.com/hlissner/doom-emacs), so doom is taking charge of setting up emacs-lsp for me.

I have been testing Emacs 28 for a few weeks now, as it was reported to improve performance on LSP integration (which was sometimes sluggish for me with rust).

I have to say I wasn't disappointed. It's now comparatively blazingly fast. I hugely recommend to give it a try if you work in emacs with LSP.

Since my distro had not packaged it yet, I installed it from guix instead (package `emacs-next`). It worked like a charm.

Don't think out of the box, but https://github.com/emacs-lsp/lsp-mode works rather well with emacs 28 from my experience. Emacs "distros" such as Spacemacs and Doomemacs both make using it a nearly "out of the box" experience.

(not supporting the editor/ide war troll, just trying to contribute some useful info here)

Yeah emacs was a revelation for me on this regard, that the idea of "Free Software" could be more than just providing souce code and license.

Emacs is actually explicitly designed to give you full power to explore, discover and modify it in order to tailor it to your needs and whims.

you'd need to prevent the people you're granting the license to from changing it in derivative works

No you don't need to. People cannot change the license of code they don't own nilly-willy, only the owner(s) of code can (which can get hairy when a codebase has multiple owners).

However a license can allow sublicensing, aka sharing the code or software under a different license, but this right must be explicitly granted by the license (or by the code owner directly), and they may be conditions attached (such as attribution)

MIT:

"[...] Permission is hereby granted,[...] to use, copy, modify, merge, publish, distribute, sublicense, [...] subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software."

GPL:

"[...] You may not copy, modify, sublicense, or distribute the Program except as expressly provided under this License."

So the MIT license grants you the right to redistribute the code under GPL (under attribution condition), but GPL does not grant you the right to redistribute the code under MIT.

(Note that all of this applies to redistributing the code or the software, not about your own internal use of the code or the software)

Aren't you worried about increasing the attack surface a lot for a machine directly accessible on the Internet?

I'm neither a network nor a security expert, but I would have thought that the router would be better running the minimum essential software only, and that the rest of that stuff would be better on other machine(s) on my local network. Especially since you have a VPN set up.

Not sure if it's the one you were thinking about, but there is a similar passage in the Slate Star Codex article "Neutral vs. Conservative: The Eternal Struggle" https://slatestarcodex.com/2017/05/01/neutral-vs-conservativ...

    The moral of the story is: if you’re against witch-hunts, and you promise to found your own little utopian community where witch-hunts will never happen, your new society will end up consisting of approximately three principled civil libertarians and seven zillion witches. It will be a terrible place to live even if witch-hunts are genuinely wrong.

Thanks for the article!

I think it's even worse actually, because this is not something that needs to be done explicitly or on purpose.

Simply training AIs with a target goal of maximizing engagement could lead to models discovering and exploiting superstimuli or superstimuli-like bias in humans.

I find the article and your comment strongly resonate with a comment I made some time ago on an article about GAN-generated faces:

I guess very soon we will be able to generate "super-attractive" (as in "superstimuli") faces for virtual personas, according to targeted demographics and purpose (advertisement, youtube videos for kids, political messages and so on).

(original comment here https://news.ycombinator.com/item?id=18310355)

Here it's not so much about the face than about the behavior and interactions, but I think the same idea hold true.

It seems to me in the near future we may well be faced with a constant exercise in self-control in the face of the multiplication of such projects.

Good point. I don't use it on mobile (I sync things between my personal and professional computer mostly) so I can't comment much on this aspect.

Joplin seems very nice indeed! But it seems to be more oriented for personal notes no? I mostly use Zotero for archiving webpages and various digital media (Zotero allows e.g. to embedded an epub as an item, or to add a link to a file stored locally).

It definitely does not solve all the issues listed in the article, but for me Zotero https://www.zotero.org has been absolutely brilliant as a "digital content organizer".

While it is "marketed" for scientists, it is a very general tool, that allows me to archive, tag, organize and search any webpage (with the option of taking a snapshot) or digital item.

I started using it for research articles, but it quickly expanded as a general bookmark organizer, then to books and even some podcast and movies archival. I use specific folders as reading/watching/listening queues.

Best of all it is FLOSS software, which is an absolute requirement for me to future-proof my use, and has an API that can be used to interact with external software. I use it for example as a document "backend" for my emacs org-mode journal (via the zotxt-emacs package).

There is an online sync service offered at a very reasonable cost (including a free tier). This is one of the very rare online services I'm paying money for. My understanding is that the sync server is open-source, but not production ready for self-hosting yet. The devs are supposedly working on it.

I strongly recommend you at least take a look.

Sure you can indeed. For me the compactness of the rpi was a plus however. Since you don't need a dedicated screen or a keyboard (recalbox interface is tailored for controllers), having only to carry the small rpi makes it more convenient and more suited to this use in my view.

Of course, if it's to play at mainly at home then I agree it's not as interesting.

If you're looking for a fun use for your shelved rpi, I can recommand installing a retro-gaming distro such as recalbox (https://www.recalbox.com/) or retropie (https://retropie.org.uk/).

I installed recalbox on a rpi3, and it runs nes/snes fine, and even some n64 games. On occasions I brought it along when visiting friends, and plugged it into the TV/projector with a couple of joysticks. It has been a blast!

All in all I'm very happy with my small on-the-go arcade machine.

The definition of "best" is often subjective and may boil down to familiarity.

On the other hand, betting on open-source tooling may be a good way to increase the chances that your "best" tools will still be available in the future.

For me, a killer feature of evil-mode is the "hybrid" state (`evil-disable-insert-state-bindings`) that allows to use emacs (aka readline) keybindings in insert mode.

When I have only a minor edit to make in the middle of typing something (like transposing two characters) it saves me the back-and-forth with the normal mode and it's really sweet.

It probably makes me an heretical freak, but I'm definitely hooked. I guess I am condemned to live the rest of my life in emacs now.

From the "About the Author" section

Marc is also one of the leads for the GNU Radio project, an open source SDR framework that is widely used in academia and defense related research. While Python is great for learning, rapid experimentation, and development, it does not lend itself well to large and computationally complex apps. GNU Radio can act as a good stepping stone for more advanced DSP apps, and a GNU Radio app or individual block is very easy to share with others.

Thanks for the recommendation, I just tried it and it seems perfect!

Contrary to your experience, I was even able to create events from thunderbird that would show up on O365. They only show up on sync however (which does not seems to be automatic, only on timer). Maybe that's the reason why it did not appear to work for you?

GitHub CLI 1.0 6 years ago

Better scripting / automation maybe? I don't have any specific example in mind right now, but being able to pipe things together with other utilities could make for interesting applications, and is easier than having to interact with a REST API.

Wow I did not expect that! Many of them are really hard to read actually!

A quick example of two positive tests side by side https://i.redd.it/llw9fmn6dxk51.jpg

Even worse, sometimes the reaction is so strong that there is a "dye stealer" effect, where the test line steals the ink from the control line. See for example here: https://i.redd.it/skdynwwi6ll51.jpg (Notice how the line is on the "wrong" side)

No wonder the misread rate is so high indeed...