> You'd be hard pressed to find a consumer-grade technology manufactured today that will still be working in 10 years.
I truly don’t think that’s true. If anything current hardware is more resilient to time than older hardware, though it is indeed much harder to fix, usually.
The software running the hardware making the hardware slower releases after releases might be another consideration, but it has nothing to do with the hardware…
I have computers from the 60s to this year and my experience is quite different; newer hardware breaks faster & is harder to repair. Everything I have from the 70s and 80s is all simply working without fail (if something is broken, it's usually just soldering a capacitor or 2); most from the mid 00s to now is broken and often not repairable. I'm talking 1000s of machines. A good exception are thinkpads from around 2010 (with the nice keyboards); they seem unbreakable.
I still have my Thinkpad from 2000 and it works flawlessly despite spending 15 years in a box in a slightly damp room (though the lid cover became sticky/tacky)
Pure anecdote here but: I own three different external HDD hard drives, all bought at different times (one of them bought used and formatted, so I don't even know how old it really is) that have all been mine since more than 10 years ago. I still use them all and they all still work fine. Obviously I also have everything on them backed up to a cloud regularly, but so far, they've been perfectly reasonable in their speed, quiet and not too warm (the opposites of these three being red flags for near failure in my experience)
I also own a secondary work/careless travel laptop that's from 2005 and that I've owned since buying it used in 2012. It too works fine, though I did have to change its hard drive once.
I knew about Collapse and Dusk OS before, but never figured out a way to easily try them (obviously I have no related knowledge). Very interesting project anyway.
I was thinking LLMs are actually an incredible compression of human knowledge- if you can swing it, a decent model and a few GPUs would be amazingly handy to help rebuild civilization.
I wonder if there has been any study of what happens to LLMs subject to bit rot, given the level of compression of facts on one hand and the “superposition of stored facts” on the other one ( I’m obviously a layman here, I don’t even have the correct vocabulary)
But that’s during training, not during inference. Also it’s more structured in where the dropout is happening. I do think that points to them being somewhat resilient but we haven’t had LLMs exist long enough for a good test
> You'd be hard pressed to find a consumer-grade technology manufactured today that will still be working in 10 years.
I truly don’t think that’s true. If anything current hardware is more resilient to time than older hardware, though it is indeed much harder to fix, usually.
The software running the hardware making the hardware slower releases after releases might be another consideration, but it has nothing to do with the hardware…
I have computers from the 60s to this year and my experience is quite different; newer hardware breaks faster & is harder to repair. Everything I have from the 70s and 80s is all simply working without fail (if something is broken, it's usually just soldering a capacitor or 2); most from the mid 00s to now is broken and often not repairable. I'm talking 1000s of machines. A good exception are thinkpads from around 2010 (with the nice keyboards); they seem unbreakable.
I still have my Thinkpad from 2000 and it works flawlessly despite spending 15 years in a box in a slightly damp room (though the lid cover became sticky/tacky)
Yep, it seems thinkpads was (is? the newest one I have is a x240) very good quality.
Pure anecdote here but: I own three different external HDD hard drives, all bought at different times (one of them bought used and formatted, so I don't even know how old it really is) that have all been mine since more than 10 years ago. I still use them all and they all still work fine. Obviously I also have everything on them backed up to a cloud regularly, but so far, they've been perfectly reasonable in their speed, quiet and not too warm (the opposites of these three being red flags for near failure in my experience)
I also own a secondary work/careless travel laptop that's from 2005 and that I've owned since buying it used in 2012. It too works fine, though I did have to change its hard drive once.
Nice article, I think base only OpenBSD or NetBSD could fill the void. I lean a bit towards NetBSD only due to its cross-compiling.
Base system of two are very usable and rather lite.
I knew about Collapse and Dusk OS before, but never figured out a way to easily try them (obviously I have no related knowledge). Very interesting project anyway.
I was thinking LLMs are actually an incredible compression of human knowledge- if you can swing it, a decent model and a few GPUs would be amazingly handy to help rebuild civilization.
I wonder if there has been any study of what happens to LLMs subject to bit rot, given the level of compression of facts on one hand and the “superposition of stored facts” on the other one ( I’m obviously a layman here, I don’t even have the correct vocabulary)
Lots of models are trained with dropout, which is kinda like bitrot at very high rates...
But that’s during training, not during inference. Also it’s more structured in where the dropout is happening. I do think that points to them being somewhat resilient but we haven’t had LLMs exist long enough for a good test
Or just get lucky and end up with a Dr. Stone [1] :)
[1]: https://en.wikipedia.org/wiki/Dr._Stone