correction because any misinfo is bad: he is indeed a US citizen (he became one in 2002) however he’s ineligible to run for president, and when he was running his startups was in the country illegally, or working counter to the conditions of his visa or something to that effect
i don’t necessarily disagree with how they’ve chosen to proceed in the US: personally, on this single issue ignoring all others I probably wouldn’t make the same decision but i don’t think it’s cut and dry (for Google)
what i think is completely fucked is they’ve stated they’re making the change for the entire world (last time i heard)… fuck you google: the world isn’t ruled by the US government… we, unlike the US, tend to follow global standards: metric, borders, place names… we have organisations that determine those. we don’t want to be part of the US (hi canada, hi greenland), we don’t want to follow you everywhere
Gulf of Mexico
worth repeating every time until the message is crystal clear
(looking at you Google you absolute boot licking assholes)
figured i’d do this in a no comment since it’s been a bit since my last, but i just downloaded and ran the 70b model on my mac and it’s slower but running fine: 15s to first word, and about half as fast generating words after that but it’s running
this matches with what i’ve experienced with other models too: very large models still run; just much much slower
i’m not sure of things when it gets up to 168b model etc, because i haven’t tried but it seems that it just can’t load the whole model at once and there’s just a lot more loading and unloading which makes it much slower
that’s true - i was running 7b and it seemed pretty much instant, so was assuming i could do much larger - turns out only 14b on a 64gb mac
i believe one of the big advancements with deepseek r1 is their method of adding the reasoning component is novel and very very efficient. i haven’t checked it out, but it could legitimately just be more efficient to run
it’s actually pretty easy to run locally as well. obviously not as easy as just downloading an app, but it’s gotten relatively straight-forward and the peace of mind is nice
check out ollama, and find an ollama UI
that’s not really how this works though… we don’t have the training data, so nobody else can recreate this from scratch exactly
*edit: adding this here because i just saw it elsewhere: https://huggingface.co/blog/open-r1
it’s possible that the grok model was trained or fine tuned somehow to help with moderation. in that case, it’s possible that things like these bullet points are somewhere up it’s context chain, or in its training data in a manner that it can relatively accurately recall
i’m not sure of the issues that you’re having, but it almost always works perfectly for me - screen mirroring, media control (streaming from device as well as remote control of existing media - even streamed from 1 device to eg a homepod and then using another device to skip etc), airdrop files and photos etc to my own device or others’ devices, even the new ability to walk away and have the transfer continue over the internet
can’t remember the last time i had an issue that wasn’t solved immediately by a retry, and even those issues are very rare
… in my experience at least
well if it’s a church raffle or charity slot machines then that’s fine
less about easy to start an EV company (honestly you can buy engines, so that’s not really the hard part; manufacturing is, as tesla found out the hard way) and more about it being easy to build an EV from almost nothing… you can ram batteries and electric motors into almost any body as you pointed out, so if a company makes junk it’s pretty easy to replace bits with whatever you like, and since electrons are electrons are electrons, your battery, motors, etc only have to kind of match
risks? it ALREADY has