iamjackg

I'm very interested in their Jailbreak evaluations: they're new to me. I might have missed previous mentions.

That seems more unlikely to me -- training is not free and takes a long time, so it would not result in "[enhancing] the usefulness of results displayed in the Gemini app" and it being "particularly helpful to our users looking for up-to-date information."

Fine-tuning, which is cheaper and faster, has been proven to not be a good solution to "teach" models new facts.

I think what's most likely here is that Gemini will have access to a form of RAG based on a database of AP articles that gets updated in real-time as new articles are published.

I've done this in the past with bugs in the grass. If I stare at a fixed point, I start seeing each individual bug moving through the grass, whereas normally they would be really hard to spot among all the fine details of the ground and grass blades.

This is what I do, the only issue is that I don't have nearly as much "range" with divergence as I do with convergence, so I have to make the pictures as small as possible when using it to line up two images (as opposed to autostereograms, which usually have a much smaller divergence offset).

It's not really a scold -- god knows I've wasted a lot of stuff for the sake of fun projects, this just happens to be past the threshold for me. Not entirely sure why. I think it might be because thermal paper feels kinda yucky: it has weird chemicals in it, it fades super quickly, and cannot be recycled. I would probably feel slightly better doing this on regular paper, although like somebody suggested I would maybe limit it to a Sunday thing.

I've had this same idea since I also own a thermal printer, but I could never get past the wastefulness of printing a piece of paper that I would then immediately throw away after reading, so I never implemented it. I still think about it periodically: there is something oddly alluring about finding a small piece of paper made specifically for me whenever I go to the kitchen in the morning. E-ink just doesn't quite feel the same.

UI-generated configs are not stored in the database, they end up in a collection of JSON files in a .storage directory inside your config directory.

Well, the various Echo devices were allegedly built as loss leaders in the hope people would use them to make orders on Amazon. This is backed by the most active open source project on GitHub, which already has extensive support for voice pipelines both with and without LLMs, and is likely priced sensibly.

A lot has changed in the open source ecosystem since commercial assistants were first launched. We have reliable open source wakeword detectors, and cheap/free LLMs can do the intent parsing, response generation, and even action calling.

The very next day?

"Integration" is just the term the HA project uses for code supporting a specific device/brand/platform. Home Assistant shows a label on each integration clarifying whether it's entirely local or cloud-based.

If you're on Android, there's an accessibility option in Firefox and Chrome that allows you to always zoom on any page, even if the website normally prevents it. On Firefox it's under Settings -> Accessibility -> Zoom on all websites.

People from (all?) visa waiver countries have their fingerprints taken when entering the US, in addition to having to submit an ESTA[0] application before arrival. Last I checked, it still asked for all your social media handles (although you're not required to provide them).

0: https://en.wikipedia.org/wiki/Electronic_System_for_Travel_A...

I'm very curious why you think that! Sincerely. These models undergo significant human-aided training where people express a preference for certain behaviours, and that is fed back into the training process: I feel like the behaviour you mention would probably be trained out pretty quickly since most people would find it unhelpful, but I'm really just guessing.

Very cool project! Just a suggestion: since you do have pre-built releases on GitHub, you should mention that in the Installation section of your readme.

Just adding a data point, here in Canada on my nVidia Shield it went down to 360p a dozen times or so, but never paused at all. I guess I got lucky.

Oh yeah, this is definitely a thing with a lot of bass-heavy EDM. I'm always a bit wary of trying new venues because I never know if they're going to have proper subwoofers or not, since some genres just do not sound right without a sound system that gives sub-bass frequencies the room they deserve.

As an example, give a quick listen to Turn Off the Lights[0]. It sounds incredibly empty and tinny on small sound systems, but add a subwoofer and you discover a whole other layer of chest-shaking bass. This stuff is made to be played live on a properly tuned sound system.

0: https://youtu.be/OgT9IpTUudc?si=pxOVShecpDbs_UCH&t=46

Ultrasound?

Take a look at https://sh.readthedocs.io/en/latest/ for a very usable solution to easily run other processes in Python. It has made my life a lot easier whenever I've had to migrate a shell script to Python.

I'm a very, very happy user of TypingMind. It's not free, but it's worth the (one-time) license cost in my opinion. It has support for multiple providers, plugins, "agents" (pre-set templates), chat grouping, import/export, automatic syncing across devices, etc.

https://github.com/TypingMind/typingmind https://www.typingmind.com/

This is definitely a problem. I mostly take a look at the various leaderboards, but there is a proliferation of fine-tuned models that makes it incredibly daunting to explore the model space. Add to that that often they're not immediately available on turn-key tools like ollama, and the friction increases even more. All this without even considering things like licenses, what kind of data has been used for fine tuning, quantization, merges, multimodal capabilities.

I would love a curated list.

The downvotes are probably because your comment was unnecessarily abrasive, saying things like "unreadable" and "abhorrent."

The feedback is valuable, but the way it's expressed leaves something to be desired.

Yeah, sorry, I think I was a bit confusing: that's exactly what I'm doing. For example, an Epic folder is laid out like this:

    EPIC-123
    ├─── user-stories
    │    └─── STORY-234
    └─── user-stories-with-summary
         └─── STORY-234-add-support-for-feature-a

Yeah, the /dev/disk/by-uuid paradigm was actually the inspiration for adding the second folder!

That's sort of what I'm doing behind the scenes, because I keep one global list of downloaded issues (they're lazily loaded when you access them) and then the folders are really only "views" into the downloaded issues. Representing identical ones across trees as symlinks is a fantastic idea though, I can't believe I didn't think of that! Thanks for the inspiration.

I wish I had known about this a month ago, when I had to go through the exact same process!

In a desperate attempt to find a less frustrating way to interact with Jira, I had the silly idea of starting a jira-as-filesystem project that uses our internal issue categorization to build a tree: directories represent issues, with files representing issue fields and subdirectories for linked issues. I ended up choosing fuse-python.

I haven't worked on it in a minute, but I was already bumping into issues (pun not intended) with the abstraction: using just the issue ID as directory name makes automation easier, but it makes it hard for humans to browse the tree, since a `ls` would just show you a bunch of inscrutable IDs. I ended up adding a parallel `<issue-type>-with-summary` directory type where the slugified summary is appended to each issue ID.

A long time ago I did some work to do exactly this in an automated fashion using ffmpeg. It wasn't perfect, but it was better than nothing. I tried going back through my bash history, and the last related entry was this command line:

    ffmpeg -i C0119.MP4 -vf vidstabtransform=interpol=no:crop=black:optzoom=0:zoom=0:smoothing=0:debug=1:input="weirdzoom.trf",unsharp=5:5:0.8:3:3:0.4 kittens-stabilized.mp4

I think the trick was to set all the stabilization parameters to 0 and crop=black to force ffmpeg to move the image around as much as necessary and zoom everything out.

EDIT: nevermind, it was more complicated than that. I actually wrote a Python script that modified the motion tracking information generated by ffmpeg to reduce the zoom amount and fit everything within a 1920x1080 frame. Man, I wish I'd added comments to this.

The https://www.reddit.com/r/ImageStabilization/ subreddit has a lot of posts in that style, but from the research I did it seems like it's mostly done manually by lining up each frame as a separate layer and then rendering an animation that adds one layer per frame.

Did you try different types of masks? I found that I felt like that during the initial phases of the titration study (where they determine the right settings for you) but once they dialed in the pressure, that feeling went away and now I actually quite enjoy it, since it makes it easier to breathe through my nose. I use a nose pillow.

I think the commenter you're replying to is just annoyed at having lost the game.

Tickling the roof of your mouth with your tongue also seems to stop incoming sneezes for a lot of people.

You read that wrong. What you're referring to is the response you get from the server in the form of callback parameters. The function signature is

    playdate.scoreboards.addScore(boardID, value, callback)

So you only send the boardID and the score. You get the player name back as part of the callback.