Damus
Open in Damus

Recent Notes

Aldin · 1d
yup, it is as you say, when set up that way. Although that caching mode is only supported on Windows. But for Linux, you have to turn it off in the bios, and then it shows up as normal nvme. I did some test. The read speed is 900MB/s, write is 150MB/s. Nothing to write home about. Latency (ping) wa...
Hynek · 1d
How do you mean that?
 m4d4m profile picture
m4d4m @m4d4m
I don't want the old system to deteriorate, crash even, and as result shift a lot of the power to the current structure of 'hodlers'.

Same as I don't want someone having the older car to experience some fatal failure and force him to jump into the Tesla-like solution. This is what for some the change may be mentally. A slower process is definitely better.

(I do not imply you are thinking like this. Just kinda also hijacking the note with my 2 cents.)
1
Hynek · 1d
Interesting point. I think that it depends on your perception of previous state. If you think it’s evil, you probably want to switch it as fast as possible. Some people can find evil even diesel cars - I don’t. But fiat system is in my eyes purely evil and therefore I wish bitcoin to voluntari...
Hynek · 1d
How do you mean that?
m4d4m profile picture
There is this Moravec's paradox.
That is: from a human perspective - easy tasks (moving objects) are the hardest to teach an AI (or make a code that will operate the HW).
And vice versa, the hardest tasks for us (thinking, writing) are easier - achievable.

I'd like it when someone challenge the old status quo.
This guy - Arvind Narayanan - (also writing the AI Snake Oil newsletter) digging in to it.

"The famous aphorism is neither true nor useful"

"Moravec’s paradox never been empirically tested. (It’s often repeated as a fact by many AI researchers, including pioneers I know and respect, but that doesn’t mean I’ll take their claims at face value!)"

"It is really a statement about what the AI community finds it worthwhile to work on. It doesn’t have any predictive power about which problems are going to be easy or hard for AI."

https://www.youtube.com/watch?v=-Ty4BXrASFU

(Kind of funny how they all pronounce the name as Moravex.)
Five · 6d
Seems like folks need nostr:npub10r8xl2njyepcw2zwv3a6dyufj4e4ajx86hz6v4ehu4gnpupxxp7stjt2p8 more than ever nostr:nevent1qqsvv6rmwpzy3z6x07aqxdz6a9kthtsz7xa5f6vcqtnn0w6487sar9gzypxhsss9z7pwp5l7kq6dz59...
 m4d4m profile picture
m4d4m @m4d4m
How does the Zapstore protect me as of now? Anyone can publish an app, even Zapstore published many GH APKs themselves instead of the original creators. Zapstore just shift trust one step, isn't it(?) - unless the npub is followed by myself and optionally/ideally this is follow done face to face I don't know if can it be trusted. Also the WoT reputation, is better than nothing, yet can't ultimately save the day. Or I am missing something. This could just happen here as well just it's not yet worth of targeting I think.
1
Zapstore · 5d
Yes you are right. What does change is the tooling we are focused on building (community-based catalogs) which is not at all the priority of big tech. Far from a finished product but WoT is really useful atm
Dune Messias · 6d
GM!
Aldin · 1w
There is this tech called intel optane. It is a fast memory storage ment to be used in RAID with an HDD. In that setup, it serves for caching, making slow HDD feel fast. It is about 10 years old; sinc...
 m4d4m profile picture
m4d4m @m4d4m
Afaik it won't help much because if model must be split between VRAM <> anything else - it's slow. And there is the KV cache for the context adding to the load. Iirc the inference, in case it doesn't using weights from VRAM, goes for RAM but this is computed on CPU - it's slow as hell (also dependent on how fast is CPU<>GPU communication). The Optane will be avoided and if not, the slowdown compared to RAM is even bigger.

Small models still do not fit in consumer HW - maybe for certain tasks & managing tool calls output but must be guided by bigger (API connected) model.

(Have no experience really, only I hope some day can offload the token consumption partially to my already existing HW, so far this is possible for simple tasks like generating the title for a conversation but not much more.)
1
Aldin · 1w
I meant the optane would be for bitcoin node. There, data is read and written on the disk constantly.