Pretty sure that’s what OP is doing, as they mention the Deck along with a kitchen chair.

If this form factor for ID isn’t well known enough for them to know about it, I doubt they received training on its existence.

Which means having this card will make them assume it’s fake by default. “That’s not what a US passport looks like!” Thereby increasing their deluded suspicions, rather than reducing them.

Like you said, it might be impossible to avoid ascribing things like intentionality to it

That’s not what I meant. When you say “it makes stuff up” you are describing how the model statistically predicts the expected output.

You know that. I know that.

That’s the asterisk. The more in-depth explanation a lot of people won’t bother getting far enough to learn about. Someone who doesn’t read that far into it, can read that same phrase and assume that we’re discussing what type of personality LLMs exhibit, that they are “liars”. But they’d be wrong. Neither of us is attributing intention to it or discussing what kind of “person” it is, in reality we’re referring to the fact that it’s “just” a really complex probability engine that can’t “know” anything.

No matter what word we use, if it is pre-existing, it will come with pre-existing meanings that are kinda right, but also not quite, requiring that everyone involved in a discussion know things that won’t be explained every time a term or phrase is used.

The language isn’t “inaccurate” between you and me because you and I know the technical definition, and therefore what aspect of LLMs is being discussed.

Terminology that is “accurate” without this context does not and cannot exist, short of coming up with completely new words.

Yes.

Who are you trying to convince?

What AI is doing is making things up.

This language also credits LLMs with an implied ability to think they don’t have.

My point is we literally can’t describe their behaviour without using language that makes it seems like they do more than they do.

So we’re just going to have to accept that discussing it will have to come with a bunch of asterisks a lot of people are going to ignore. And which many will actively try to hide in an effort to hype up the possibility that this tech is a stepping stone to AGI.

Obviusly.

And like hallucinations, it’s undesired behavior that proponents off LLMs will need to “fix” (a practical impossibility as far as I’m concerned, like unbaking a cake).

But how would you use words to explain the phenomenon?

“LLMs hallucinate and lie” is probably the shortest description that most people will be able to grasp.

Yup. The way the article titled itself isn’t helping.

Fun fact, you can felt dog hair but not cat hair

You can absolutely felt cat hair.

I don’t even need needles to get my cats fur to solidify into a ball, I just massage a bunch of it like a snowball.

Seems like it’s a technical term, a bit like “hallucination”.

It refers to when an LLM will in some way try to deceive or manipulate the user interacting with it.

There’s hallucination, when a model “genuinely” claims something untrue is true.

This is about how a model might lie, even though the “chain of thought” shows it “knows” better.

It’s just yet another reason the output of LLMs are suspect and unreliable.

Triple AAA, maybe.

Indie Studios have genuine fans, who don’t need the latest GPU, CPU and RAM to keep buying their games.

The gems aren’t hard to find. Good games have fans singing them praises from the rooftops.

I could give you an almost endless list if you asked for good games, and they wouldn’t all be old.

I’m referencing this.

I’m antitheist, personally.

We have the power of god and anime on our side.

The actual one.

How did you install kopia? What system are you on?

I’m not yet sure about that

It needs to be running, if it is, it will follow the policy. Systemd can start it with the system, but you can also start it some other way. Or you can execute snapshots without it constantly running, via cron/script. It’s up to you.

Abrubt power off shouldn’t damage the drives, but it can absolutely corrupt files that are being written, and even entirely delete things in the write cache.

Depends. If you are running it as a service that starts with the system (sudo sysemctl enable kopia should work with most install methods, as kopia comes with a systemd service you only need to enable) then yes, it will use its own scheduler.

If you want to use your own scheduling, you’d use anything that can execute a command on a schedule.

The main advantages of Kopia, are speed and destination flexibility.

The off-site storage does not need to have Kopia installed. It can be a mounted network location, an FTP server. Whatever. A generic cloud storage bucket like Backblaze B2.

That’s why just a router with and external drive hooked up is able to suffice.

For all of these, you can connect multiple Kopia instances to that same destination, and each client can browse backups, restore from them, and backup their own files to the destination. It even performs file deduplication across different source device. All while that destination device or service, has no access to your encrypted files.

With borg, you need something like a Pi that can have borg installed. (You can also do this with Kopia, in which case the Kopia instance on the destination device is also able to manage the backups).

Kopia also beats borg and restic in speed. My daily backups typically complete within a minute or two. I used to use Duplicati, with which it was common for it to take up to an hour. When it started regularly taking more than an hour, I switched to Kopia.

Kopia is not the fastest for initial backup. The speed of this varies depending on destination type. It does not compress by default, but you can enable almost any type of compression you want. No, what it is fastest at is updating backups. If there is nothing to update, it does not take forever for it to figure that out. Kopia does it in seconds.

https://www.intel.com/content/www/us/en/support/articles/000098345/graphics.html

Looks like you can add VP9, and need to remove VC1

Neat! That was a dealbreaker back when I last tried it.

I use Jellyfin with the Symfonium mobile client.

Navidrome is popular but does not support multi-tags for some fields, like artists.

1. There is no “special” benefit to a pre-built NAS. They have convenient software but there is nothing exceptional about them. They’re just computers with storage drive slots. Using a bunch of external drives via a USB hub would be fine. But is that your only expansion option on the system you have? Access speeds via USB, especially if using a hub, won’t be ideal. It’ll certainly work, though. You can also get enclosures to put full size HDDs in, which can connect to an existing system.

2. RAID is still the way to go, but since you don’t need much storage, I’d start with RAID 1, not 5. 5 will require a rebuild with a new drive if something goes wrong, while RAID 1 will work with 2 drives and give you complete mirroring. Since you intend to have a “local” backup copy anyway, why not just skip that and use RAID 1? It’s literally the same thing, except it’ll actually provide uptime in case of failure, unlike a backup drive or raid 5.

3. So RAID 5 plus a local backup, plus another offsite? This is overkill IMO. (Not the offsite backup that’s good. But raid+local copy. Just use two drives and mirror them using whatever you prefer.) In your place, I think I’d go with BTRFS in raid1c2 mode. This is like raid1, in that with two drives, you only get the capacity of one drive. But, the “c2” means that each data block is mirrored to two drives. With more than two drives, you can expand storage. (With three 2TB drives you’d get 3TB) You don’t get as much available storage as with raid5, but you get expandability, which you normally don’t with raid1. And you get uptime in case of failure without an array rebuild (though for this you must mount the volume with the “degraded” option, unlike actual raid using mdadm). You also get filesystem snapshots.

4. You intend to do this manually? That is fine. My current solution is a second NAS at my dad’s home, to which my system is backed up daily using Kopia. Kopia deduplicates and compresses the backups, efficiently keeping versions up to two years back. The simplest version of this would be a router that can host an FTP server using an external drive in its usb port. This way you could automate off-site backup and have it happen more frequently. Asus routers can do this, and even come with free dynamic DNS and automatic https with letsenrypt. You literally just plug it into WAN somewhere, and you’ll be able to back up to it over the internet.

Finally, just some mentions.

MDADM, is what you’d use to create a software RAID array.

BTRFS has built-in multi-device storage, of which only single, raid0, and raid1 are stable. Do not use the raid5 and 6 modes. While named raid, the modes differ from actual raid. BTRFS is able to convert from one mode to another, and can add drives in any mode (though will need to “balance” the drives after changes, to make additional capacity available). It is also able to evict drives. It will not auto-mount a volume after drive failure, and requires the “degraded” option be added.

Mergerfs can be used to merge filesystems to expand storage non-destructively. It is able to arbitrarily combine volumes of any type, to combine their capacity. This way, it can for example be used to expand a raid1 array by combining it with a single disk, or another raid1 array, or whatever else. This can be done temporarily, as the combined volume can also be disassembled non-destructively, with each file simply remaining on whatever drive they were on.