zellyn.com
He/Him.
Programmer. Square. Former YouTuber. Vaguely South African. ≡𝐺𝑂
Born 331.12 ppm CO₂
hachyderm.io/@zellyn
85 posts
83 followers
46 following
Regular Contributor
Active Commenter
comment in response to
post
I always assumed everyone agreed Russia was aggressively invading and violating another nation's sovereignty, and that it was absolutely wrong, and the kind of thing that should not be tolerated in a modern world, and something somebody should absolutely do something about.
Somebody else.
comment in response to
post
Then you simply need to do that for every region you're in, and make sure you have appropriate Terraform maps set up. Because the Terraform resources want the fake names…
comment in response to
post
Go to "Resource Access Manager" (RAM) in AWS console, and click "Resource Access Manager" in the left nav, and it'll show you the mapping:
comment in response to
post
Pretty sure this is sarcasm, but Poe's Law rules everything around me 😂
Battlestar was a postmodern gem in that it felt like it made sense in context, and it was only when you stepped back that you thought, "Wait, wasn't 2/3 of the entire last season hanging on a burning question you just dropped?"
comment in response to
post
I believe the writer actually has a plan. But considering how angry I still am over Lost and especially Battlestar Galactica, I definitely see your point!
To me, Severance is so well done, they could cancel it right now forever, and I'd still feel it had been worth watching.
comment in response to
post
You would probably like Ezra Klein's recent interview with Kara Swisher. So much shade thrown at techbro billionaires.
comment in response to
post
I understand if y'all don't want to tangle with money. One of my favorite things about hearing you on podcasts was appreciating how many of y'all are true believers, which I ❤️ so, so much!
comment in response to
post
As the feed grows, the fun of running a labeller will get dwarfed by the pain of trying to run Twitter-scale infra.
With my scheme, it's a revenue source (or can be dialed back to break-even), and also becomes a platform where the best algorithms can earn revenue for maintainers.
comment in response to
post
Postscript: I included a couple of random tasks I personally have found helpful (out of many). Nicholas Carlini put the argument that LLMs are profoundly useful much better, and more thoughtfully: nicholas.carlini.com/writing/2024...
comment in response to
post
Again, there are a thousand critiques you can level against AI and those pushing it. But let's, please, stop claiming they're just glorified Markov Chains.
comment in response to
post
Second: that they are just mimics or parrots, and can't “reason”:
* Making predictors bigger and having apparent reasoning pop out is still mysterious, and was wholly unexpected
* I defy anyone to claim their definition of “reasoning” a decade ago would not be 70% covered by current LLMs
comment in response to
post
First: that they're not useful.
Anyone who says this hasn't tried them. I've:
* built a spline editor I needed for diagramming, in minutes
* discussed the reason other "Zellyn"s are usually female and culturally Chinese
* had thoughtful conversations improving my understanding of parts of math
comment in response to
post
Here's what Claude says…
/ducks
comment in response to
post
It's been so long since YOLO was popular, seems like a lifetime ago…
comment in response to
post
I will never forget watching the scroll at the end of Cry Freedom.
My dad took advantage of a family stay in Hawaii in 1990 to rent the movie, which was banned in South Africa.
comment in response to
post
That article reads to me more like, “Please don't count Amazon AI out of the game. We have clever solutions to your worrying problems, Coming Soon™”
comment in response to
post
Is that vaguely describing a system like the one DeepMind published about a year ago:
> AlphaGeometry is a neuro-symbolic system made up of a neural language model and a symbolic deduction engine, which work together to find proofs for complex geometry theorems.
deepmind.google/discover/blo...
comment in response to
post
“Avoid, at all costs, arriving at a scenario where the ground-up rewrite starts to look attractive.”
So good.
comment in response to
post
I did have some trouble getting Claude to:
(a) create an artifact rather than just some javascript code
(b) not switch over to React when mistakenly asking it to create a "Component" instead of an "Artifact"
comment in response to
post
@simonwillison.net thought you might like this LLM test case
comment in response to
post
Mistral Small 3 is far more… interesting
comment in response to
post
Claude is amazing, as always:
comment in response to
post
(Apologies for linking to 𝕏)
comment in response to
post
Like all awesome Internet Facts, I assume this is actually false. Still delicious.
x.com/Brillianto70...
comment in response to
post
Ezra Klein's talk with James Pogue was incredibly enlightening to me. Also, terrifying.
www.nytimes.com/2025/01/28/o...
comment in response to
post
I think you meant "best of duck"
/ducks
comment in response to
post
Dunno, ask the folks who were stuck at Uber for like 50 years…
comment in response to
post
Oh I have _so_ many more! 😂
comment in response to
post
The management of persistent identity across successive otherwise purely functional renders of declarations to infra is also an essential complexity. That's why TF has state and everything else either uses TF or has something akin to TF state.
comment in response to
post
I see the declarative/imperative split as essential complexity of the space. Perhaps in the future when the Right Answer has emerged, it'll include a principled handoff between the two as fundamental.
comment in response to
post
This looks great. Hope my questions didn't come off as critical: just trying to understand the boundaries of what Massdriver handles.
comment in response to
post
I mean, if @bcantrill.bsky.social and @ahl.bsky.social start actually explaining things at the start of Oxide and Friends episodes, then yes
comment in response to
post
bsky.app/profile/zell...