Christian Kothe @christiankothe@qoto.org

@gvy_dvpont maybe v2.0 should have a motor that makes it flutter

Learned a new term of art in CG (old passion of mine) while visiting the Harz Mountains in Germany: Brocken spectre. To me that raises a question. Does your spectral renderer correctly reproduce Brocken spectres?

@crude2refined yeah that doesn't surprise me much; for one, feature flags like bias or learnable scale and so on (but really the other 20 parameters baked into the 5 or 6 different layer types too) are easily overlooked hidden bits enabled by default in the off-the-shelf layers of DNN packages. And moreover, when transformers first burst onto the scene they appeared as one fairly complex piece of machinery with all sorts of design choices locked in by just one team (of course after plenty of testing and tuning). There's no way that all those choices could have been optimal even for their own dataset let alone everything that they've now been applied to, and of course the greater community has been finding quite a lot of big & small improvements over the years (eg long-context attention etc) and are still finding them. I'd say it's not even obvious that anything in GPT-2 or later would be a part of the ultimate sequence model that we'd have in 20+ years.

@badlogic A couple months ago I felt a similar itch to get into electronics & Arduino etc, and ended up watching the EECS course by Anant Agarwal (one of the originators of edX). Felt good learning the "hardcore" version & helps w/ confidence when dealing with power electronics.

@pervognsen Tried to replace mine after E key disintegrated the other week (biggest beef for me personally was having to recharge it every few days), but mistakenly got the newer S variant (not compatible with the unifying receiver) and then the Mac variant (never rebuy electronics in anger). I used to be a decade-long evangelist of the HP wireless elite (v1 or v2) before, but went for the cheaper HP wireless this time around. Keys are a tad firm compared to MX and elite (which we still have at the office), but I guess for $14 it wasn't a waste.

@simon some of these have in common that they're not really the low-level substrate that AWS is known for but some grafted-on buzzwordy me-too stuff (e.g., Cloud9, AppStream, Finspace Dataset Browser, QLDB, Workdocs, Mobile Hub) or some now rather dated tech stacks (OpsWorks). But I always viewed CodeCommit as potentially good for resilience since you don't want to have your AWS stack's CI depend on some other hotshot startup's high level Rube Goldberg machine that may go into maintenance mode at the slighest disturbance at Azure, GCP or AWS itself, for that matter, so that's a bit shocking to me (fortunately we never used that).

@zeh For a moment I thought this was waxing about AWS development (Lambdas triggered off S3 triggers initiated by their own object uploads... cloud accounts and their owners' wallets going up in smoke and glitter in a beautiful runaway reaction).

@pervognsen @wolf480pl @pkhuong Got it, makes sense (FWIW, EFS might also be an interesting option for a cloud-backed fully managed file system, though mounting that from outside the cloud might take some config work)

@pervognsen @wolf480pl @pkhuong Unsolicited comment; I'd feel a bit queasy about building something of that fine granularity on top of S3 (there's also some minimum size requirement for intermediate parts IIRC), but DynamoDB would seem like a fair fit if performance/parallelism/no-servers-to-manage mattered (good performance and there are some basic transactional features).

@zeux Guilty as charged. Now the q is what's the 2024 way of panning on PC? Like, what's the first thing that someone would try in a CAD tool they just opened for the first time, on Win and Mac?

@pervognsen I switched to Linux as my primary system in 2020 and each time I have to go back to increasingly new versions of Windows to use some software there, I am shocked at how broken the experience has become since it peaked a couple years ago. From spurious bluescreens to every single app coming up with a few-second delay (notepad, calculator, paint, ...), to a sluggish startup, to "we update you whenever we feel like it" (and reset your settings while at it), to IntelliJ feeling several times slower on the same repos, to menu and UI animations feeling glitchy and half-baked, to a sluggish startup etc, etc...

Valve should hire the guy who made that HL3 fan film, today. https://www.dsogaming.com/videotrailer-news/this-is-what-half-life-3-should-look-like-if-it-ever-comes-out/

@mcc @dpiponi Was going to say, in yak shaving it's subshaves all the way down.

Quite impressive work on LLM interpretability, with applications.
https://transformer-circuits.pub/2024/scaling-monosemanticity/index.html

@rooster@chaosfem.tw That could have featured in the opening scene in Forrest Gump FWIW.

@grumpygamer there's your new character name ;)

@tdietterich I'm actually impressed that they find time for a book at all amid fiercest competition with OpenAI, and now from a bit of an underdog position. There could also be more on how this can be synthesized into a probabilistic programming language like e.g. numpyro (sampling, variational inference).