Thanks for the info. I have not really tested Seedvault myself so this is all good to know.
Ironically, one of the main reasons I switched to GrapheneOS was because Google’s backups were so frustrating and I was hoping Seedvault would be more comprehensive.
What’s wrong with Seedvault?
But here’s the really funky bit. If you ask Claude how it got the correct answer of 95, it will apparently tell you, “I added the ones (6+9=15), carried the 1, then added the 10s (3+5+1=9), resulting in 95.” But that actually only reflects common answers in its training data as to how the sum might be completed, as opposed to what it actually did.
This is not surprising. LLMs are not designed to have any introspection capabilities.
Introspection could probably be tacked onto existing architectures in a few different ways, but as far as I know nobody’s done it yet. It will be interesting to see how that might change LLM behavior.
I’m not sure what the exact model is, but it’s probably from the Performa or Power Mac 5000 or 6000 series. It’s low-res so it’s hard to read, but the text next to the floppy drive says “PowerPC”, referring to the CPU family used in Macs in that era.
The screen looks like Mac OS 8. It’s so low-rest that it’s kind of hard to tell, but the menu bar at the top of the screen is clearly from Mac OS. Could be 7.5, but I’m guessing 8 since that’s what’s shown in the web browser.
I think the left screen is showing Windows. Again, super low-res, but those look like Windows 95/98’s blue window title bars and gray task bar at the bottom.
Same.
That was probably the intention. X-Files was at its height of popularity around this time (assuming 1997 by the Mac model and OS 8).
It’s insane that this is even legal.
I’ve never actually tried it, but I think you could use BTRFS subvolumes to multiboot without partitioning the physical space.
And then maybe even use deduplication across subvolumes?
BTRFS can work across multiple disks much like ZFS. It supports RAID 0/1/10 but I can’t tell you about performance relative to ZFS.
Just be sure you do NOT use BTRFS’s RAID5/6. It’s notoriously buggy and even the official docs warn that it is only for testing/development purposes. See https://btrfs.readthedocs.io/en/latest/btrfs-man5.html#raid56-status-and-recommended-practices
Edit: Another interesting thing to note between the two file systems is deduplication. ZFS supports automatic deduplication (although it requires a lot of memory). BTRFS supports deduplication but does not have built-in automatic dedup. You can use external tools to perform either file-level or block-level deduplication on BTRFS volumes: https://btrfs.readthedocs.io/en/latest/Deduplication.html
Good advice!
This is also available with BTRFS. Personally I am leveraging this feature via Snapper, simply because it was the default on OpenSuse and was good enough that I never bothered looking into alternatives. I’ve heard good things about Timeshift, too.
This has saved my butt a couple times. I’ll never go back to a filesystem that doesn’t support snapshots.
I really liked ZFS when I used it many years ago, but eventually I decided to move to BTRFS since it has built-in kernel support. I miss RAIDZ, though. :(
If you got a problem, reinstall and do the same stuff again, you’ll almost certainly get the same problem again
Sure, but nobody’s likely to do that. If I wiped my system now, I doubt I could get it back to exactly the same state if I tried. There are way too many moving parts. There are changes I’ve forgotten I ever applied, or only applied accidentally. And there are things I’d do differently if I had the chance to start over (like installing something via a different one of the half-dozen-or-so methods of installing packages on my distro).
For example, I have Docker installed because I once thought a problem I had might have been Podman-specific. Turned out it was not. But I never did the surgery necessary to fully excise Docker. I probably won’t bother unless and until there is a practical reason to.
Also worth mentioning: you might still need to add the “most recent visit” column under the View menu. And if you dare to actually load any of those pages, they’ll move all the way to the top, and will not remain in their original location. It’s really annoying.
Whisper is open source. GPT-2 was, too.
Absolutely this. Phones are the primary device for Gen Z. Phone use doesn’t develop tech skills because there’s barely anything you can do with the phones. This is particularly true with iOS, but still applies to Android.
Even as an IT administrator, there’s hardly anything I can do when troubleshooting phone problems. Oh, push notifications aren’t going through? Well, there are no useful logs or anything for me to look at, so…cool. It makes me crazy how little visibility I have into anything on iPhones or iPads. And nobody manages “Android” in general; at best they manage like two specific models of one specific brand (usually Samsung or Google). It’s impossible to manage arbitrary Android phones because there’s so little standardization and so little control over the software in the general case.
btrfs’s RAID features are not production-ready, and at this point I doubt they ever will be. See:
https://en.wikipedia.org/wiki/Btrfs#Implemented_but_not_recommended_for_production_use
https://wiki.archlinux.org/title/Btrfs
https://www.phoronix.com/news/Btrfs-Warning-RAID5-RAID6
ZFS is definitely more robust.
Is this legit? This is the first time I’ve heard of human neurons used for such a purpose. Kind of surprised that’s legal. Instinctively, I feel like a “human brain organoid” is close enough to a human that you cannot wave away the potential for consciousness so easily. At what point does something like this deserve human rights?
I notice that the paper is published in Frontiers, the same journal that let the notorious AI-generated giant-rat-testicles image get published. They are not highly regarded in general.
DuckDuckGo is an easy first step. It’s free, publicly available, and familiar to anyone who is used to Google. Results are sourced largely from Bing, so there is second-hand rot, but IMHO there was a tipping point in 2023 where DDG’s results became generally more useful than Google’s or Bing’s. (That’s my personal experience; YMMV.) And they’re not putting half-assed AI implementations front and center (though they have some experimental features you can play with if you want).
If you want something AI-driven, Perplexity.ai is pretty good. Bing Chat is worth looking at, but last I checked it was still too hallucinatory to use for general search, and the UI is awful.
I’ve been using Kagi for a while now and I find its quick summaries (which are not displayed by default for web searches) much, much better than this. For example, here’s what Kagi’s “quick answer” feature gives me with this search term:
Room for improvement, sure, but it’s not hallucinating anything, and it cites its sources. That’s the bare minimum anyone should tolerate, and yet most of the stuff out there falls wayyyyy short.
Something like for-jay-yo.
From https://forgejo.org/faq/ :
Forgejo (pronounced /forˈd͡ʒe.jo/) is inspired by forĝejo, the Esperanto word for forge.
“Smart” may as well be synonymous with “unpredictable”. I don’t need my computer to be smart. I need it to be predictable, consistent, and undemanding.
I’ve been using Kagi for a while, so I’ll post a few quick thoughts I had after reading the article, linked blog, and mastodon thread.
The one thing in the blog post I strongly disagree with is her statement that the summarizer is “the same old AI bullshit”. I think they just assumed that without actually testing it. The summarizer is fantastic, and is a great example of the right way to use LLMs. Its output comes entirely from the URL or file you specify. It does not hallucinate. You can ask it follow-up questions about the document, and again, its replies are limited in scope to what’s actually in that document. If you ask it something out of scope it’ll tell you that it can’t find that information in the document. This is great because it’s using the LLM for what LLMs are actually good for — complex language parsing — and not for what they’re bad for, like reasoning or information storage/retrieval. It’s actually quite difficult to misuse the summarizer. It’s straightforward and effective. This is Kagi’s killer feature, IMO.
I can’t speak as highly of its search-integrated AI features like FastGPT. They mostly take information from the contents of the first few search results, but they also seem to “fill in the blanks” more than I’d like. Sometimes it gives me information that is simply not in the sources that it cites. It’s not as bad as using ChatGPT (which hallucinates all day every day, and if it ever tries to cite source is hallucinates those, too) but it needs improvement.
That said, Kagi doesn’t shove the AI down your throat like you might think reading the blog post. These are separate features that need to be explicitly activated. Plain searches don’t return results from the LLMs. If you want that, you need to explicitly go to the assistant or trigger the “quick answer” feature on the search results page. EDIT: I just realized that this is not totally true as of a few weeks ago, when they added a feature to automatically trigger the “quick answer” blurb for search queries ending in a question mark. Perhaps that’s why Lori felt like it was over-emphasized.
Anyway, back to the summarizer, here is an example of it in action. I gave it the URL of the Mastodon post. I think this is an excellent result. I also asked it an unrelated followup question so you can see that it limits itself to the information in the URL. It will not make shit up!
The summarizer lets me download conversations in markdown form, so I’ll just paste it right in here so you can see.
Exploring https://hackers.town/@lori/112255132348604770
Assistant:
Key points:
Query:
What’s the capital of North Dakota?
Assistant:
The knowledge provided does not contain any information about the capital of North Dakota. The context is focused on an email exchange between the author and the CEO of Kagi search engine regarding criticism of the company.
I used to use Filen for this, but it never worked very well. The file provider path it returned to Keepass2android was only temporary, so it would break periodically. Did Filen change how that works?
I eventually started using Syncthing instead. I connect to my home wi-fi often enough that it’s never too far out of sync with my home PC. And since it’s a local file, there’s no issue with using absolute paths.