realtime api + elevenlabs but llms will be diversified based on persona moving forward. Using chatgpt/gemini as baseline model, we feel prompting has limitation
As time goes on, the satire of Eddington becomes increasingly poignant. I'm amazed you could come up with an analogy like this in regards to holding accountable people granted lethal force in a democratic society.
Eh, not for laptops - I say as someone who switched to Linux from windows in past year.
I have spent a decent few days to get long battery life on Linux (fedora), with sleep hibernate + encryption. And I am still thinking that the Linux scheduler is not correctly using Intel's pcore/ecore on 13th gen correctly.
If you have an Nvidia GPU you're generally going to need to edit the systemd services and change some kernel settings. This is a real pain point to be honest and it should be easier than it is (usually not too bad tbh)
If you want I can try to help you debug it. I don't have a fedora system but I can spin up a VM or nspawn to try to match your environment if you want
I just got a lunar lake laptop and in CachyOS you can just enable either scx_lavd or scx_bpfland from the kernel settings. I use them both: bpfland guarantees that the active application runs smoothly even if you compile code in the background, and lavd focuses on energy saving a bit more. They both understand how to use the P and E cores: especially the lavd scheduler puts the active app to a P core and all the background apps to the E cores.
It is a bit of a pain to set up though, especially for a laptop - I just switched to fedora+ gnome, and going through configuring the power settings to allow for suspend-then-hibernate was annoying.
Figuring out the luks + page file + hibernate resume configuration is non intuitive, and is only viable for me to figure out due to my Linux based day job.
Probably could have gone bazzite and had things just work, but I need a Linux dev box locally and was not sure about dev on bazzite.
Yes, with the direct conclusion from that being tl;dr in theory OPs explanation could mitigate RAM, in practice, it’s worse
(Source: I maintain an app integrated with llama.cpp, in practice no one likes 1 tkn/s generation times that you get from swapping, and honestly MoE makes RAM situation worse because in practice, model developers have servers and batch inference and multiple GPUs wired together. They are more than happy to increase the resting RAM budget and use even more parameters, limiting the active experts is about inference speed from that lens, not anything else)
Long term storage, along the lines of tape replacement/supplement?
There are plenty of things that need to be archived in a basement and never read unless the more readily available forms get corrupted.
Having the ability to say as long as the item exists the data exists is valuable, especially with not having to worry about degradation (which happens with tapes/flash/hard drives)
My impression of this is that it is partially a tax policy issue.
Dividends are taxed differently and higher than capital gains. So given a choice between a stock buyback and a dividend, often a buyback makes more sense.
The issue is that the Democrats 'compromised' on the last funding bill - basically had an agreement that some programs would stay funded.
Then the trump administration said "funding allocations from Congress are limits, and we do not have to spend that much on these programs if there is not a need"
reply