Promiseland

[Fetian]

It would be fantastic in that context, honestly

[Fetian]

One of my superpowers is that I have crazy good timing when getting into something -- I'll get into something and shortly afterward that thing will be renewed, or a sequel will be announced, or it'll come off hiatus, or. I got into the Dark Tower series shortly before Wolves of the Calla was published, for example, when the previous book had come out six years before that.

So last week I was thinking about albums or soundtracks that would be neat to have on a record, now that I have a record player, and I thought the Over the Garden Wall soundtrack would be a good one. I did some searching to find out if it had ever come out on LP and if so, how much they were going for -- yes and very expensive, respectively. So I dug up the company that produced them to see if they still sold them and/or if there were going to be any rereleases, and found out that they had been doing semi-regular limited releases that sell out extremely fast, and also the same day I was looking into all this, this had been tweeted:

So obviously I set an alarm to get on here and buy it when it drops. Yesterday there were more tweets that the aim is to just be able to have it continuously in stock, so if it runs out they'll just print more and sell those. So trying to get a jump on it wasn't completely necessary but also I didn't want to take any chances

There were of course some technical difficulties but the long and short of it is that I did get an order in and when they start shipping I will get a gorgeous LP of the Over the Garden Wall soundtrack.

[Fetian]

It appears to be being very nice out today and I'm trying to decide if I should head over to dad's today or if I should put it off a week

[Fetian]

Obviously walking to dad's didn't happen, instead I have spent today zoning out in front of the computer. In my defence waking up early enough to snag that record, unnecessary as it turned out to be, did mean less sleep last night

Plan now is to wait for ash to come home with food, eat the food, then take a shower, and then either work on a project with ash or work on ket myth stuff. And then closer to evening show ash the last three episodes of game changer

[Fetian]

I just want to know why recovering from trauma has to be so damn inconvenient

[Fetian]

Screenshot_2023-03-10 r discordapp - Comment by u ReallyAmused on ”Features Discord has planned in 2015”.png

In a thread about discord features, reply is from a person who works for discord

Do... do discord devs really not know how bad the search feature is...? People have been complaining about it for years, have they ... never seen the complaints? The support tickets and the bug reports and the feature requests? Do they not look at those at all?

[Fetian]

Anyway, there's Talk happening that discord is going to start scraping conversations to feed to AI, which has some extremely troubling implications, and when confronted on it discord employees made a statement that they're not going to train Clyde specifically on user data without user permission and said nothing at all about the other features (automod, summaries, avatar-editing), which is a very loud silence.

[Fetian]

Oh I should see about getting like a record of carnival music

[Fetian]

Anyway, there's Talk happening that discord is going to start scraping conversations to feed to AI, which has some extremely troubling implications, and when confronted on it discord employees made a statement that they're not going to train Clyde specifically on user data without user permission and said nothing at all about the other features (automod, summaries, avatar-editing), which is a very loud silence.

from same employee:

We don't intend to use your images or art to train generative models. We're not in the business of training generative models, and even if we were, using the data that users have uploaded to our platform to train these models is a non-starter for us as well. It would be a huge violation of user trust to train generative models on data they have uploaded to the platform without their consent. So, no, there isn't a future where we will use your art will be used without you knowing to train a model. However, if this changes in the future for whatever reason, we will communicate this clearly, and in no uncertain terms.

Of course, model training is nuanced. For example, we have a spam detection model, and when you report spam to us on the platform, we use this to train the model to be better at detecting spam and classifying spam reports. This is how every single spam-filter you've ever used works. So, we can't outright say we're not "training models on data you provide us" - but I think we can all agree that training a spam filter on reports of spam to us, is something we'd all want to reduce spam on the platform in general.

We're still finalizing details of the rollout and have been monitoring community feedback around this. We'll announce more as we get closer to launch.

In the case of a summarization model, for example, chat summaries, this is also not trained on message data. It uses a pre-existing model, in order to perform summarization. You can think of this really as a "fancy compression algorithm" rather than a "generative model". In terms of what the computer is doing, it's no different than compressing an image. The image resizer isn't remembering what images it's resized previously. It's doing an operation on data given an algorithm, and then forgetting about the image entirely when it's done with its job of resizing.

We're working on updates to the privacy policy to have more clear and precise verbiage around this. Give us some time to get those out of the door.

Still not totally reassuring -- for understandable reasons -- but sounds like there's discussion happening within the company about all this

[Fetian]

Definitely a lot of pollen going around