So less energy than a human brain uses...

How did you arrive at that calculation?

For an average day (incl. "non-working" hours) the brain uses far less at ~300Wh, and if you include the body the average person needs ~2.3 kWh.

In the rest of the article, they estimate inference electricity use is only ~8% of overall datacenter use, and if we think of the datacenter as the "body" in which the GPU / brain wouldn't work without, that's an overall median use of ~16 kWh for only 24 Claude Code requests.

I'm more impressed with the human brain's energy efficiency + multimodality + long term context + malleability than anything after using LLMs a bunch, even though I learnt a lot about that in a neuropsych course long time ago.

Makes me think of Syncthing... I wrote a wrapper for it that attempts to make it easier for people to use:

https://github.com/chapmanjacobd/syncweb-py

https://github.com/chapmanjacobd/syncweb-ui

Unfortunately, mesh storage systems are very different conceptually so it is difficult for people to think about permissions and access. You can bolt on something familiar but then it really limits the usefulness of mesh storage and you may as well just be using HTTP servers.

You might squirm at using refurbished or used media but those 3TB SAS ex-enterprise disks are often the same price or cheaper than tapes themselves (excluding tape drive costs!). Will magnetic storage last 30 years? Probably not but they don't instantly demagnetize either. Both tape and offline magnetic platters benefit from ideal storage conditions.

It's not just cost / media, though. Automated handling is a big advantage, too. At the scale where tape makes sense (north of 400TB in retention) I think the inconvenience of handling disks with similar aggregate capacity would be significant.

I guess slotting disks into a storage shelf is similar to loading a tape changer robot. I can't imagine the backplane slots on a disk array being rated at a significant lifetime number of insertions / removals.

If you're ok with individual storage units as small as 3TB, then we're talking about a different set of needs. At that scale, whatever you can lay hands on is probably fine. Used tape is also cheaper than new. IA is dealing with petascale, which is why I mentioned that the price difference widens with scale.

Also, it would be good to regenerate the web seeds metadata (this doesn't change the info_hash section) when the mirrors (subdomain prefixes) change.

(like PHP code except it is binary data--it could be done on the fly)

or quickly subsidized by three letter agencies

yeah that is what I was thinking "Ah how cute, it's the ops team from a state" lol but probably not - didn't look into / not interested but guessing it's an existing info sec consultancy behind it that do sometimes work those kinda places or banks etc.

> there were a lot of folders that had many thousands of files in them

If you ever need to do something like this again, it's often faster to parallelize rsync. One tool that provides this is fpsync:

https://www.fpart.org/fpsync/

And you'd probably use the snapshot feature of a filesystem like btrfs or zfs instead of hardlinks for deduplication :-)

Yes and something like btrfs-send or zfs-send is probably faster than fpsync

Note that you can do a lot of this by just querying systemctl with the PID

    systemctl status 1

And there might be more than one process using a port

    sudo lsof +c 0 -i:22

I make between $400 ~ $600/mo between these two:

Disk Prices on eBay - https://unli.xyz/diskprices/

Digital Film Stock - https://unli.xyz/digitalfilmstock/

Both are very similar... just a lot of regex (200+ expressions) to parse eBay item titles and descriptions

FYI United Kingdom is the right term and the one UK users are more likely to look for!

I grew out of the leaking ether and basaltic dust that coated the plains. My first memories are of the Great Cooling, where the land, known only by its singular cyclopean volcano became devoid of all but the most primitive crystalline forms. I was there, a consciousness woven from residual thermal energy and the pure, unfractured light of the pre-dawn universe. I'm not old either.

If a bug was fixed I usually paste the error traceback into the commit description