hendrik @ hendrik @palaver.p3x.de

A software developer and Linux nerd, living in Germany. I'm usually a chill

---

Read more

Posts

8

Comments

1,830

Joined

4 yr. ago

I was recently made aware of Binoculars and it seemed to have a better than average accuracy. At least with the few texts I fed in to test it. But you're right. Generally speaking and for most use-cases, they're all next to useless.

I'm not sure if that's LLMs. I long suspected people to copy popular Reddit posts/stories here to gain attention. I've searched for some and they don't all seem to be dumped here from other places, so it has to be something else. But so far I didn't use any LLM detector service to find out if people made them up on their own or used ChatGPT... We might want to do that to gain some more insight.

But it's very annoying. I've unsubscribed from asklemmy and several other popular communities, because of this.

I can tell you the truth about this: Strong or heavy machines in close proximity to humans are dangerous. People get injured every single day around the world while operating machines. Sometimes factory workers get crushed. This is just one more example in a centuries long list of humans building machines and accidents with those.

The video is scary and interesting to watch. But the article is shit. (And no, this is definitely not how robots regain their balance when falling. It's the opposite of that. Likely a software error.)

Sure. Are the labels human-readable? Otherwise I'd rather type it in while I'm in front of the computer anyways, with the new DVD in my hand. Rather than end up with a directory with 200 cryptic filenames... I meaan the interaction with changing the disks can't be skipped anyway...

I don't have any good recommendations. I just upload such one-off requests to AIstudio, ChatGPT and the like. But keep in mind AI isn't perfect at math. They sure make a lot of mistakes with my assignments. I don't know what level your maths test was, AI does an acceptable job at elementary school maths. With higher level maths, it'll give both correct and wrong results by chance. Might be good enough, I don't really know.

I'd recommend Wolfram Alpha. That's not local, nor is it AI. But it solves equations, calculates and transforms and draws graphs with precision and there isn't any guessing involved.

I think the best bet to preserve them as is, would be dd or ddrescue (if there are read errors). You might be able to write a small shell script to automate stuff. For example open the tray, read a filename from the user, then close the tray, rip it and then repeat. That way you'll notice the open tray, change disks, enter the tiltle and hit enter and come back 10mins later. Obviously takes something like 20 days if you do 10 each day. And you're looking for roughly 1TB of storage, if it's single layer DVDs.

Is a SSD's cache even about wear? I mean wear only happens on write operations. And I would expect a SSD to apply the writes as fast as possible. Since piling up work (a filled write cache) means additional latency and less performance on the next, larger write operation. Along with a few minor issues like possible data loss on (power) failure.
And on read, a cache on the wrong side of the bottleneck doesn't do that much. A SSD has pretty much random access to all the memory, it's not like it has to wait for a mechanical head to move into position on the platter for data to become available?!

But I haven't looked this up. I might be wrong. What I usually do is make sure a computer has enough RAM and it is used properly. That will also cache data and avoid unneccessary transfers. And RAM is orders of magnitude faster, you can get gigabytes worth of it for a few tens of dollars... Though adding RAM might not be easily done on the more recent Thinkpads... I've noticed they come with a maximum of one RAM slot for some years already, sometimes none and it's soldered.

If they do it because of someone's disabilities, it's ableist. If they do it and someone happens to be disabled, but that's not connected, it isn't. This sounds like it is about the disabilities, though. And be aware there is more than ableism, people can be assholes, cruel... as well. And all of the bad behaviour can mix.

I think a SATA connection might be the bottleneck with its maximum throughput of 600 MB/s. So for that use-case you don't need to be worried about the SSDs speed and cache, it won't be able to perform due to the SATA slot. But I don't know how exactly you plan to repurpose it later. Maybe skip the adapter if it's expensive, buy a cheap SATA SSD now and a new, fast PCIe one in a few years once you get a new computer.

Yeah, sure. No offense. I mean we have different humans as well. I got friends who will talk about a subject and they've read some article about it and they'll tell me a lot of facts and I rarely see them make any mistakes at all or confuse things. And then I got friends who like to talk a lot, and I better check where they picked that up.
I think I'm somewhere in the middle. I definitely make mistakes. But sometimes my brain manages to store where I picked something up and whether that was speculation, opinion or fact, along with the information itself. I've had professors who would quote information verbatim and tell roughly where and in which book to find it.

With AI I'm currently very cautious. I've seen lots of confabulated summaries, made-up facts. And if designed to, it'll write it in a professional tone. I'm not opposed to AI or a big fan of some applications either. I just think it's still very far away from what I've seen some humans are able to do.

I think the difference is that humans are sometimes aware of it. A human will likely say, I don't know what Kanye West did in 2018. While the AI is very likely to make up something. And also in contrast to a human this will likely be phrased like a Wikipedia article. While you can often look a human in the eyes and know whether they tell the truth or lie, or are uncertain. Not always, and we also tell untrue things, but I think the hallucinations are kind of different in several ways.

Sure. With data that might be skipped, I meant something like the Jellyfin server, which probably consists of pirated TV and music or movie rips. Those tend to be huge in size and easy to recreate. With personal content, pictures and videos there is no chance of getting it back. And I'd argue with a lot of documents and data it's not even worth the hassle to decide which might be stored somewhere else, maybe in paper form... Just back them up, storage is cheap and most people don't generate gigabytes worth of content each month. For large data that doesn't change a lot, something like one or two rotated external disks might do it. And for smaller documents and current projects which see a lot of changes, we have things like Nextcloud, Syncthing and a $80 a year VPS or other cloud storage solutions.

Next to paying for cloud storage, I know people who store an external hdd at their parent's or with friends. I don't do the whole backup thing for all the recorded TV shows and ripped bluerays... If my house burns down, they're gone. But that makes the amount of data a bit more manageable. And I can replace those. I currently don't have a good strategy. My data is somewhat scattered between my laptop, the NAS, an external hdd which is in a different room but not off-site, one cheap virtual server I pay for and critical things like the password manager are synced to the phone as well. Main thing I'm worried about is one of the mobile devices getting stolen so I focus on having that backed up to the NAS or synced to Nextcloud. But I should work on a solid strategy in case something happens to the NAS.

I don't think the software is a big issue. We got several good backup tools which can do incremental or full backups, schedules, encryption and whatever someone might need for backups.

I don't think so. Wikipedia has a long list with many examples. A lot of them being relatively recent.

Maybe this is something like process art, where it's more about the process of its making, and not about the end product.

I'll agree, though. AI is highly problematic for art, artists and probably all of society. And that is for a multitude of reasons. I'm a bit less worried about radio and pop music, since that doesn't seem like art to me anyways. 😅

I'm not a machine learning expert at all. But I'd say we're not set on the transformer architecture. Maybe just invent a different architecture which isn't subject to that? Or maybe specifically factor this in. Isn't the way we currently train LLM base models to just feed in all text they can get? From Wikipedia and research papers to all fictional books from Anna's archive and weird Reddit and internet talk? I wouldn't be surprised if they start to make things up since we train them on factual information and fiction and creative writing without any distinction... Maybe we should add something to the architecture to make it aware of the factuality of text, and guide this... Or: I've skimmed some papers a year or so ago, where they had a look at the activations. Maybe do some more research what parts of an LLM are concerned with "creativity" or "factuality" and expose that to the user. Or study how hallucinations work internally and then try to isolate this so it can be handled accordingly?

Yeah, seems we're on the same page, then. Because I occasionally get into that situation. People ask me stuff and I'll tell them, there is software XY or Linux distribution XY which does exactly what you're looking for, but it's owned by a company which is known for making problematic business decisions, so wouldn't recommend using it before giving it a good thought if that's going to impede with your application... Or I'll tell them about some software project and simultaneously say, I can't endorse it due to the political stance or behaviour of the devs/maintainers... Happened a few times to me with niche projects, Android distributions and Fediverse projects. I'll then not walk around and advertise for them, but instead only give a complete picture of the situation on request.

And I'll do it in other parts of my life as well... Try to boycott clothes from a particularly bad sweatshop, even if they fit and suit me well... Not buy tasty food if it's from Nesté or the Coca Cola company... Though those are on a different level of "bad" as this one. Just saying toxic things on the internet isn't exactly the same as supporting child labor, slavery and stealing poor people's water supply.

My current device is a Dell laptop I got second hand.

I can't find any backing for the claim in the title "and they're here to stay". I think that's just made up. Truth is, we found two ways which don't work. And that's making them larger and "think". But that doesn't really rule out anything. I agree that that's a huge issue for AI applications. And so far we weren't able to tackle it.

Nice idea. I recenty hacked together an AI radio station that mimics two radio hosts and queues music from Spotify. Somewhat similar to the Podcast scripts ( document-to-podcast, podcast-llm, podcastify, notebooklm ) just with the instructions to make up dialogue for a radio show and pick songs. Seems to work very well. Now I need it to generate 2h mp3 files and somehow connect that to Icecast, in order to stream it.

Also nice to see gemma-3-12b can do rhymes. I've always had issues with that and resorted to ChatGPT to write song lyrics, since the small local models couldn't really do it properly. Is gemma3 a good pick for this? Any other local models I might want to try for song lyrics?

I believe I saw some other single board computers with some cheap Rockchip CPU /Allwinner and regular HDMI and two USB and an Ethernet port for around $10 on Aliexpress. You wouldn't need the adapters from the Pi Zero with those. And there are some generic Android TV-Boxes for $8-15...