HNHacker News
TopNewBestAskShowJobs

MezzoDelCammin

586 karma · joined November 23, 2021

submissionscomments
MezzoDelCammin··on What happens to our breath when we type, tap, scroll
<mild storyline spoiler alert> except if You're in the void just outside the main continent </spoiler>
MezzoDelCammin··on A market mystery: Why do capers come in such tiny jars?
life hack : use the other end of the spoon (or any other utensil, ideally one that has a slightly concave handle).
MezzoDelCammin··on The Effects of Early Relational Trauma (2001) [pdf]
Sorry You had to go through that...

But if we're doing an AMA : mind writing a bit more about the recovery? You mentioned humor as a coping strategy, so that makes me think You've probably been through some kind of introspection / therapy. Mind writing a bit more on what worked / didn't in Your case?

MezzoDelCammin··on North Yorkshire Council to phase out apostrophe use on street signs
I can't believe nobody has yet mentioned the obligatory relevant XKCD:

https://xkcd.com/327/

MezzoDelCammin··on India's TikTok Ban Is a Cautionary Tale for the U.S.
Could You please be more clear on how the last sentence relates to the first?

If You think Prohibition (of alcohol) is somehow related to prohibition of TikTok, then corollary would be "correct answer to alcohol threat is better alcohol".

MezzoDelCammin··on Don't fuck with paste
yep. Quite the WTF
MezzoDelCammin··on Bookish Diversions: Do Audiobooks Count?
Seems we're fairly divided here between "audiobooks are for fiction" and "they work great as learning resource".

How about a slight tangent / challenge: Try to give a few examples of "which" audiobooks/podcasts/lectures are in Your own personal Top 3 or so (and maybe why).

My personal kickoff (I'm mostly in the fiction / "I need narrative" camp):

1. "The Poet's Corner" by John Lithgow (narrated by himself). This book probably converted me to even prefer audiobooks over regular paper / ebooks. Lithgow is a wonderful actor and in case of poetry, interpretation matters as much as the text itself. And the selection of different actors he brings in would be hard to beat even if it was done by a major Hollywood studio.

2. Bobiverse by Dennis E. Taylor, narrated by Ray Porter. If You like SF and haven't yet tried anything that's been narrated by Porter, please do yourself a favor and check his name out. I'll wait. The list is worth it.

3. Postwar by Tony Judt, narrated by Ralph Cosham. This one convinced me that what I need in an audiobooks is a narrative, but it does not necessarily need to be fiction. I'm a bity of a history nerd and reading one of the major works of 20th century in an audio form was the final straw that convinced me that audiobooks work great even for some non-fiction cases.

Basically, whenever I see I have a chunk of 30+ minutes of time when I "can" get away with running on autopilot and devoting part of my attention to listening (in my case - whenever I am cycling / running training, commutes, highway driving etc.), I now have the option of reading at the same time.

MezzoDelCammin··on Notes on paper (2023)
OK OK, let's get serious. My personal favorite : Lamy 2000 fine nib inked with Rohrer & Klinger Dokumentus. Most times to complement it there's at least one Parker 51 as a quick note taking pen / highlighter (puts out way more ink, so some cheaper papers don't really work with it) and one or two Montblancs (mostly inked with Rohrers Salix) just for the fun of it.

I haven't quite settled down on paper though. Rhodia wire-bound dotted A5 has the best quality paper, but I sorely miss the flexibility of free leaf notes. Latest discovery - Muji's A5 binder. It's a Japanese 20-holes-per-A5 spacing, so fairly rare in the EU. Also, the Muji's default paper isn't great for some wetter fountain pens... but hopefully I'll figure some way to make it work.

MezzoDelCammin··on Show HN: Multi-monitor KVM using just a USB switch
Wouldn't count on that. Thunderbolt is USB-C just visually. It's probably better to think of it as PCIe that coincidentally has the shape of a USB connector.
MezzoDelCammin··on Can You Live Code a Church Organ? [video]
Anyone else reminded of Knuth's "Royalties, use of" in TAOCP?
MezzoDelCammin··on Tell HN: Russia has started blocking OpenVPN/WireGuard connections
Damn. So sorry, mate! I was born not far from a Soviet base back before the curtain fall, but nothing like Donbas.

Fingers crossed and some contributions to the cause already done. In the meantime : Slava Ukraini!

MezzoDelCammin··on YouTube started slowing video buffer with adblock enabled
OK, this has been brewing on for a while, so time for a Plan B. Since I don't have enough karma do do a full AskHN, here's to hijacking a comment thread:

Question : What's the best solution for someone mainly watching STEM channels and various video essayists (Contrapoints, Stories Of Old, various art critics)?

Personally, Youtube Premium feels like paying for a protection racket. So far I'm inclined towards paying for Nebula. Any other ideas anyone?

MezzoDelCammin··on A structured note-taking app for personal use
Thanks for bringing this one up. It's almost strange to see a civilized discussion about something that clearly polarizes a lot of people in the thread. Makes me wonder exactly what kind of social dynamic results in a bunch of people actually talking through their differences, instead of the these days usual trench warfare...
MezzoDelCammin··on CUDA vs. ROCm: A case study
True, bitcoin miners have probably figured out quite a few things to which I'm fairly oblivious. I always regarded that movement as a complete waste of computing power. Turns out at least some of their stuff may turn out to be useful for LLMs...

As for the specs, I went new Threadripper way simply because that's AFAIK where You'll get the best balance of PCIe 5.0 lanes (for the memory bandwidth between the GPUs) and power.

I see how You could spec this out with some used Xeon, but for one I really wouldn't like the headache and another, You're probably getting PCIe 2.0/3.0 tops (if You're trying to cut costs that much).

That Supermicro You mention seems to be an interesting compromise. The MB is selling second hand for about ~700EUR in the EU. The Threadripper that goes into it... don't know, can't find them used, but guessing something under 1k EUR as well.

Interesting price, but then on top of the risk of using old parts, instead of partitioning PCIe 5.0 (and maybe having the option to go 4.0 and double the count, since Your GPUs are PCIe 4.0 anyway), You're now splitting straight 4.0 lanes to Your GPUs. Half the bandwidth, but it may end up not mattering all that much. Given that most LLMs are memory size / memory bandwidth limited, I probably wouldn't like to risk this. But that's just me...

MezzoDelCammin··on CUDA vs. ROCm: A case study
$9k for three of those? I'm affraid Your math aint mathing.

7-slot motherboards are IMO Threadripper teritory. The CPU alone will set You back at least $1k5, MB to go along with that is another $1k. Not including the stack of DDR5 that You'll likely throw into that and the PSU / case / coolers to feed it power and evacuate the heat from this monstrosity.

That's at least $3000 for the base system "before" You add any GPUs.

And at some point I'd like to read someone more enlightened to comment on what the memory bandwidth is likely to be and how exactly will You distribute the PCIe lanes...

MezzoDelCammin··on More people should write (2012)
I just did look. 19k karma is indeed nothing to sneeze at.

Difference is though that in the HN comments it's someone else who picks the topic and time.

Much as I appreciate the discussion here, occasionally I think there's some value in just doing a writeup on a topic. Or, a memory dump. If for no other reason, than for the therapeutic value of cleaning up the mind... And forcing the random musings that make sense in the head through the filter of written language just to see what actually makes sense.

MezzoDelCammin··on As YouTube declares war on ad blockers, Google sponsors ad blocking conference
I kinda thought about this for a few months now. My problem with YouTube these days isn't ads. It's content quality.

It's probably just my personal preference, but what I'm looking for is long form. Lectures, classes, essays, analysis... This sort of stuff.

And it seems that from my viewpoint we're now in some sort of race to the bottom. There seems to be more and more of a push towards short Tiktok style stuff. And I hate that with passion.

My current solution is probably subscribing to Nebula, not YouTube Premium.

MezzoDelCammin··on As YouTube declares war on ad blockers, Google sponsors ad blocking conference
"Ads is the sole source of revenue" is plainly wrong, unless You're maybe just beginning. Looking at some breakdowns (strangely, quite a few channels publish them) the ad revenue seems to be just a minor slice these days.

The actual income is either from sponsorship, Patreon, merch, or just plain "advertising yourself" (seemingly the case of many photo/videography channels).

It's not too much of a stretch to say YouTube ads are detrimental to any of these revenues.

MezzoDelCammin··on The importance of handwriting is now better understood
The answer will probably be "it depends". On smoothness, fountain pens have the edge in that mechanically, a fountain pen is just a very smooth ball of metal that's being constantly lubricated by a controlled leak of ink through capilary action. With some decent enginnering and manufacturing (caveat : both of these cost money), this will be hard to beat when it comes to pure friction.

When it comes to practicality... Let's say one does not buy a fountain pen to be practical ;-) These things are fairly maintenance heavy and prone to accidents. There's a reason they have been superseded by several other pen types these days.

Fountain pen is a tool from a different era. One that excels at putting ink on paper in a very comfortable / stylish / craftsman manner...

MezzoDelCammin··on The importance of handwriting is now better understood
oh, just You wait until You discover the wonderful new old world of fountain pens and inks ;-)
MezzoDelCammin··on Amsterdam preparing to lower speed limit to 30 kilometers per hour
if anyone is interested in a comparison, Paris has had this for some months now. The main idea was to slow / limit traffic through the residential areas (which is most of the city center). There has been an initial wave of skepticism, but things have calmed down somewhat as people seem to have gotten used to it.

Personal (anecdotal) experience : -cars now go about as fast as some bikes. Makes it easier for cyclists to merge / left turn through even some of the bussier intersections.

-during the bussier parts of the days, speed limit is useless. Nobody is doing 50 on Monday afternoon and it seems that the 30 helps with fluidity / merging. Which (at least in my experience) means that though the traffic jams still exist, they're moving somewhat faster

MezzoDelCammin··on Exllamav2: Inference library for running LLMs locally on consumer-class GPUs
Ah, sorry, I was talking about the money I would pay after the expense deduction of 50%. So, 4090 at $1600 would cost me about $800 of my own money in the end. That being said, I'm writing from the EU, so all these numbers are really just ballparks where I take 1EUR==1USD for the sake of simplicity.

Thanks for the rental tips though. I'll take a look. It's indeed more effective in the beginning...

MezzoDelCammin··on Exllamav2: Inference library for running LLMs locally on consumer-class GPUs
yeah, that's the thing. The AI/ML is something I'd like to play with / learn / experiment on, but it's only tangential to my current paying job at this point in time (I'm basically a freelancer).

"Expense" in my context means that I'll deduce about 50% off of the invoice come taxes time. I'm OK paying $400 of my own money for an AMD card, I'd think twice, but probably accept coughing up $800 for a 4090, but $2k is definitely over the budget at this stage.

Thanks for sharing the spreadsheet with Your benchmarks though. You're going into my "HN commenters to watch" list ;-)

MezzoDelCammin··on Exllamav2: Inference library for running LLMs locally on consumer-class GPUs
yeah, a used 3090 would be the most likely candidate if my next build was something for personal use only. Thing is, I'm probably going to expense it. Which limits me to new components.

And I'm not in too much of a rush (weeks / short months). Can afford to wait for a month / two to see how the ROCm support pans out on the AMD side. 60% performance for <50% price is a compelling argument. If the ML support is decent...

Though the 4090 is somewhat expensive, I might end up taking it anyway just for the simple reason that it's the only 24GB card with CUDA.

MezzoDelCammin··on Exllamav2: Inference library for running LLMs locally on consumer-class GPUs
Interesting bit at the end of the text:

2023-09-13: Preliminary ROCm support added

Makes me curious how will the RTX4090/3090 compare with something 7900-ish

MezzoDelCammin··on Exllamav2: Inference library for running LLMs locally on consumer-class GPUs
4090Ti? AFAIK that has never been released
MezzoDelCammin··on Oscilloscope watch ships after 10 years on Kickstarter
Cade's bike fit is generally OK, but IMO he's lately been way too click-baity with the videos and trying to tailor his texts to the youtube algorithm ("10 things ...", etc.). Maybe it's just me, but I find it somewhat annoying to sift through.

Probably wouldn't hurt to second-check these days.

My $.02 : Cam Nichols for bike fit tips (and a wonderful aussie accent :-) and Dylan Johnson for science checks and backwards cap comments ;-)

MezzoDelCammin··on Oscilloscope watch ships after 10 years on Kickstarter
(1) in my case, "bike fit" is more of a process than a one time thing. I keep tweaking it a bit through the year as I train (mainly saddle fore/aft, but handlebar height / angle as well as extra wrap / gel pads sometimes)

(2) the distance alone is just part of the job. I do about 5.000km a year just on the 2x10km a day commute. That's got little to do with mobility / intensity though. And does little to posture. For me,I kinda regret now doing only a couple of 200km+/day low intensity rides this season. This would have revealed the problems much earlier...

(3) 1200km is Paris-Brest-Paris. The time limit for that one is 90h, so 86 is "barely just made it". But I went full bikepacking on this one. Took way too much gear, which slowed me down a lot in the climbs. Also, couldn't use my aerobars in the groups, which put way more stress on my wrists than I thought it would. In the end, it's the combination of vibration / limited options for hand position change / too much weight (fault of weak core muscles / posture) and the long continuous repetitive stress that does the damage.

MezzoDelCammin··on Oscilloscope watch ships after 10 years on Kickstarter
second the "check Your bike fit". In my opinion, the checklist is:

1. check the bike fit (probably the main problem)

2. add some mobility / core training (at some point if You bike long distance out of the blue without much training, some bits in Your body will simply break / strain no matter how good Your bike fit is)

3. in case of problems, look up the terms "cyclist palsy" and "ulnar / median nerve flossing exercises". It should help recover a bit quicker.

All that being said, I'm typing this with a pretty severely tingling median nerves in both hands. 1200km in 86h without much training during the season and with an improperly adjusted saddle (didn't mark the position before packing the bike for transport). Yeah, do as I say, not as I do.

MezzoDelCammin··on Ghostfolio: Open-source wealth management software
And have You managed do somehow automatize it in the end?
← PreviousPage 3 of 8Next →