Hasn't been updated since 2020, but Tim Dettmer's guide [0] is pretty much the gold standard for optimizing what to buy for which area of DL/ML you're interested in. The pricing has changed thanks to GPU prices coming back down to earth a bit, but what to look out/how much ram you need for which task hasn't. Check out the "TL;DR advice" section then scroll back up for detail info on why and common misconceptions. For the tips on a RAID/NAS setup alongside it, just head to the datahoarders subreddit and their FAQ.
I'm so tired of these "infinite canvases" being proposed as solutions for organization without any second thought to how you'd retrieve any information. Minimaps are a possibility, but there are no shortage of ways to get ideas down both linearly and nonlinearly.
I thought the Zettelkasten hype would show more people that when it comes to organization and notetaking we need better ways to get things out than in. And when comes to the idea of "automatically linking" notes and ranking concepts within those notes based on importance or value there's no such thing as "one size fits all." Ideas and innovations in that area, maintenance, search, and surfacing relevant (forgotten) notes are far more interesting problems.
Love how temp accounts can post partisan hackery (and post a comment noting they have an obvious agenda), so we get to hear all the old, lame conspiracy theories from twitter on hn.
Meta only focused on biomedical information. This is explicitly based on Microsoft Academic Graph which was all encompassing. They did remove patents, tag papers with fewer topics/subtopics, and don't index as many conferences, but they did add a couple of nice things like ORCID. [0]
I would've linked the Nature article and study[0][1], but of course it's paywalled. The blog does however have a video of Emily Jones finding new lines with it, and I've seen her put in some stunning times in both iracing and GT Sport.
EDIT: Oh and here's [2] a (long) video of the agents in action...their racing lines and aggression are genuinely impressive.
Nothing about this requires blockchain. There are plenty of Twitter archive sites like nitter.net that are not only open source, but also have plenty[0] of distributed instances on good old "web 2"
> "Nitter can only archive (and update) data Twitter allows them to access."
That's exactly what you'd expect from a Twitter archive site. The author's point was that this is the same tech you'd need for a Twitter replacement, and this doesn't need blockchain or web3.
Reminds me of the comment on HN from a few months back "On web2 music artists only get 18% of the revenue. Spotify gets 33%. The music industry takes 49%. Web3 flips the script." You don't need blockchain or web3 to fix that either - centralised solutions like bandcamp.com already give the artist 80-90%.
If enough people don't like the changes in an open source system, they can get together and fix it or fork it. I see a number of others comments mentioning the already working (non web3) solution Mastodon - and that's a great example of this in practice.
To look at it another way, how would web3 actually help here? As far as I can tell it would only make this example worse - a more user hostile UI (faffing around with wallets and keys and whatnot), less decentralised (everything going through one API to interact with the blockchain), more resistant to change (forking blockchains is a big deal and only tends to happen when rich people at the top of the pyramid don't like something), more expensive (gas fees), less privacy (your wallet address being essentially an undeletable cookie), more likely to incentivise divisiveness and conflict (flamewars driven by the expectation of personal profit rather than just likes), etc. etc.
I wasn't aware ETH Zürich was considered the Ivy League!
But all jokes aside I think you should give these professionals a little more credit--they also factored that possibility into their experimental design and included tests for "useless" token transfers [0]. While parrots would often transfer tokens to empty compartments (for when they knew their neighbor was missing and lacked tokens), these same parrots did not when there was a completely empty partition [1]. Further even the article notes they didn't just hand out tokens automatically or willy-nilly, but were more or less willing depending on the bond.
Personally I think the most interesting thing about the study is not that non-mammalian animals have the capacity for altruism or prosocial behaviors, but that Blue Headed Macaws (despite being extremely intelligent as well) were not willing and the difference is that their general populations form smaller (though equally cohesive) flocks compared to AGPs.
EDIT: and the reason why they used tokens vs direct walnut transfer is that these birds were trained in a prior experiment (with an equally interesting premise) [2]
As someone who's gotten both X2100 and X300 mods from Xyte (a much smoother process than what originally was dropping 1k+ into a random person's account on a forum and hoping you actually got your custom motherboard months later) I just love that they're still trying to expand on this concept and recognize the lasting fervor around OG Thinkpads.
Personally I just want a Ryzen-powered mainboard, but figure there's sourcing reasons for why they're sticking to this intel generation.
Regularly check on meta, saw the announcement, and figured they'd change the name so they can transfer the .org and .ai names to the main company. Didn't expect them to just shutdown the team altogether. With their May/June updates (where they started to track and index beyond just papers) I thought they were actually onto something.
[0] https://timdettmers.com/2020/09/07/which-gpu-for-deep-learni...