Webtorrent.io is probably the best thing I’ve come to so far with p2p webrtc. Really amazing what came out of PeerCDN.
4,299 karma · joined April 23, 2018
#2 It looks like its time to pivot. Find a niche which you can exploit (like bio it seems) and conquer it. Google Docs looked like it was going to dominate everything and no one had to do anything else in that space, yet Notion comes along and is now valued at $2B.
Obviously you'd probably have it saved on your computer, but this may mitigate / in some instances stop it.
I'd still say its a bit sketchy, but it makes sense in context of the everyday CS student.
Nuro for example had this open house that exclusively invited Harvard/MIT students (i snuck in) and I'm sure there's a lot of similar opportunities that are open to kids in these colleges.
In the long term, I'm sure it'll fade away, but starting out matters as to where you will end up (whether you go to X startup or get to work for X VC)
Despite how much success/failure/I learn I endure, college name matters... a lot. I snuck into Nuro's open house last year and found they exclusively invited Harvard/Stanford/MIT students (I can say this because I was a minor at the time of signing their NDAs)
College name matters, especially when you're starting out (which can impact where you evenetually end up... whether its working for a startup or working at a cool VC firm) In fact most of the time, the people who've said: "don't go to college" are the same people who go to Cornell, UPenn, and Stanford and probably don't realize how much of a network/presitgie gives them to get them where they are now.
If K8s supported compose scripts out of the box (not Kompose) that'd basically make Swarm unnecessary (at least for me)
1. Swarm is dead in the water. No big releases/development afaik recently
2. Swarm for me has been a disaster because after a couple of days some of my nodes slowly start failing (although they’re perfectly normal) and I have to manually remove each node from the swarm, join them, and start everything up again. I think this might be because of some WireGuard incompatibility, but the strange thing is that it works for a week sometimes and other times just a few hours
3. Lack of GPU support
"Few-Shot Learning with a Large-Scale Language Model" makes more sense.
Even with their robot hand paper, they titled it along the lines of "we solved a rubrix cube" not "a robot hand manipulated the cube and solved it"
https://github.com/shawwn is doing some work in the GPT-2 space including using TPUs instead -- which has given him pretty good results.
/s (although sometimes it's true)
If you wanted to generate poems with GPT-2, you'd need to have a lot of poems to fine-tune GPT-2 to get reasonable results.
With GPT-3, you use few-shot learning instead (without the need to do gradient updates with each example)
The paper is long and filled with how it stacks with models like Grover and T5 and it does well... given that this is a 175 B param model (relative to Grover/T5's 1.5/11B param models). This shows that even with these huge models, smaller models can outperform them in certain instances with lesser param models.
Also I think they did a good job with explaning the ethics and morals around what models like these mean / what biases this has.
Obviously there's a power issue (afaik)and several other issues, but ARM support is hard for many projects. I know you can build docker containers for all architectures via qemu/buildx, but for stuff like Longhorn by Rancher for example or even Envoy (I believe) dosn't support ARM for some weird reasons I haven't gone enough into.