Not somebody who knows a lot about this stuff, as I’m a bit of an AI Luddite, but I know just enough to answer this!

“Tokens” are essentially just a unit of work – instead of interacting directly with the user’s input, the model first “tokenizes” the user’s input, simplifying it down into a unit which the actual ML model can process more efficiently. The model then spits out a token or series of tokens as a response, which are then expanded back into text or whatever the output of the model is.

I think tokens are used because most models use them, and use them in a similar way, so they’re the lowest-level common unit of work where you can compare across devices and models.

Agreed! I’m just not sure TOPS is the right metric for a CPU, due to how different the CPU data pipeline is than a GPU. Bubbly/clear instruction streams are one thing, but the majority type of instruction in a calculation also effects how many instructions can be run on each clock cycle pretty significantly, whereas in matrix-optimized silicon its a lot more fair to generalize over a bulk workload.

Generally, I think its fundamentally challenging to generate a generally applicable single number to represent CPU performance across different workloads.

I mean, sure, but largely GPU-based TOPS isn’t that good a comparison with a CPU+GPU mixture. Most tasks can’t be parallelized that well, so comparing TOPS between an APU and a TPU/GPU is not apples to apples (heh).

I just came across the lines in the OpenSuse 42 .bashrc in to connect to palm pilots today…what a flashback.

Same question, on vanilla android.

They could have at least renamed it to Radeon Operational Compute method or something…

Yeah heliboard is the only one I’ve found that is actually usable on a day to day. Just wish the autocorrect was better, other than that no complaints.

Yeah, I think the battery thing OP pointed out makes more sense than the power argument. The Z1 extreme used in other handhelds is based on the 8840HS iirc, anf its at least one generation newer than the basis for the steam decks somewhat custom silicon.

The Deck processor is 4 Zen 2 CPU cores and 8 RDNA 2 GPU CUs, while the 8840HS is 8 Zen 4 CPU cores plus 12 RDNA 3 graphics CUs. It’s going to be wildly more powerful. The 8745H actually has the same CPU and iGPU configuration as the 8840HS – not even close to steam deck specs.

AntennaPod is better than it has any right to be – on a modern device, it’s super smooth.

Isn’t that going to be ruinously expensive to host an instance for? Video is expensive in terms of storage and bandwidth.

Yikes! Best of luck.

Oh boy! How new is your kernel? If you’re on older than probably 6.11, the Strix driver might not be implemented in it yet.

Obsidian is not FOSS, but you can switch to it for now because the whole idea is that it’s just a folder of markdown files. I recommend shopping around by pointing each app at the same markdown folder, so you can see your same notes without having to worry about complex migration. Being able to look at all your notes gives you a better idea of what will suit you.

Also, I recommend Pandoc for translating between document formats. It’s not not absolutely perfect, but it is wildly good at dealing with the complex problem of translating.

The simplest thing you can use, IMO, is Marktext. It’s basically Notepad for markdown – no file manager, no special features on top of the markdown syntax, etc. Beyond that you start getting into what features you want on top, at which point you really do just have to test them out for your use case.

As far as options go, you have basically two options as far as systems go:

• A built-in sync in the program between the desktop and phone version of the same app (i.e. obsidian, Joplin)
• Use a tool like syncthing for sync between devices, which allows you to use any app you want for actual note-taking, and allows you to use different apps on phone vs. desktop. I do the latter, and use Zettlr on desktop (more document than note-centric) and Markor on android. The issue with the former is that a great desktop app might not have the best phone version. Also, the apps that do sync typically use an internal database that you can export as a folder of markdown files (i.e. Joplin), and don’t actually just look at a folder of files. This makes testing out different editors kinda hard, unfortunately.

The other wierd variable is that some apps are literally just a WYSIWYG markdown editor (Marktext, etc.), whereas most of them are markdown editors with Other Stuff On Top™ (Obsidian, Zettlr, LogSeq, etc.). Not all apps implement the same flavor of markdown (which can be maddening, but you can use pandoc to change markdown flavor), but if you rely on a specific app’s special flavor of garnish on top of markdown, it becomes harder to switch to another app in the future if you prefer its functionality or UI. Just something to keep in mind.

For me personally, one of the make or break traits is a good table creator. Making tables by hand in Markdown is a maddening, so having a GUI way to do it makes a huge difference if you end up needing to make a lot of tables. That is really hard to find because it is hard to automate Markdown table formatting in a foolproof way. As far as I know, the table plugin in Obsidian is the best way to do that by far at the moment. The Zettlr devs are working hard on rewriting theirs from scratch to be way more robust, but that is WIP.

tl;dr Just pick a Markdown editor, and you can shop from there as long as they store their files in a simple folder.

Yeah, kobo does too. I assumed it was a proprietary flavor which was pretty locked down, is that not the case?

I vaguely remember there being a FOSS OS you can put on Kobos, can you also do that on Boox?

What issues do people have? I’ve never had/seen these issues.

I mean they do have a point: the framework that the game is targeting is DX11, so if it looks bad it is (broadly) because of an issue in translating DX11 instructions to Vulkan…

Level1 has looked at the B580 on Linux specifically: https://www.youtube.com/watch?v=Tv0o6505JAc

I think most of the issues with games not working should be the same between windows and Linux driver versions, and HardwareUnboxed has done some pretty exhaustive testing recently on the “maturity” of the drivers by testing a couple hundred games for obvious driver problems.

You’re allowed to like gimmicks!

What makes them a gimmick IMO is that they’re sold as “this will change your life and the way you work”, but really it’s just that a subset of the audience for the gimmickless product thinks they’re kinda neat.

Sure the threat model is different, I’m just saying it’s still a single point of failure.