Christopher Swenson (Cupcake)
swenson.io
Christopher Swenson (Cupcake)
@swenson.io
I like cats, sorting, cryptography, math, and programming. Software engineer usually. Wrote a cryptography book and a sorting library. he/him
The Prime Rib Directive
December 27, 2025 at 1:14 AM
I think the current approach to LLMs of just throwing more energy at them until they seem better is irresponsible. What if we could reshape the solution space (in this case, programming language) instead?

I think that is an interesting thing to investigate at least.
December 22, 2025 at 10:27 AM
It’s more that people do use the LLMs, and I think it would be interesting to investigate programming languages from that perspective. What if changing something about programming languages makes LLMs more effective or say, more energy efficient? Won’t know unless we look.
December 22, 2025 at 10:23 AM
I think a simpler language could be faster to compile and end up faster than C if done right, like FORTRAN.
December 21, 2025 at 10:48 PM
A question: what are your goals for the language?

If I were designing a language in 2025, I’d probably design a language with the goals of: easy for the compiler to optimize, easy for an LLM to work with (less complex syntax and maybe fewer features? Not sure), and maybe easy parallelism?
December 21, 2025 at 10:35 PM
“Juice isn’t worth the squeeze”

I think you meant “the Diet Coke isn’t worth the can”

😎
December 19, 2025 at 7:21 PM
I agree that it is not okay but perhaps necessary.

I did see that. It rings a little hollow tbh, like when people try to justify supporting HP content by saying they’ll just not directly give money to JKR. Or eating at Chick-fil-A sometimes.

And look, I’m not perfect. There’s no winning.
December 19, 2025 at 4:04 AM
Thanks. That is a little clearer.

And I get it. Nothing is easy and none of the choices are objectively correct, like The Good Place taught us.
December 19, 2025 at 2:40 AM
It was not clear to me. You did explicitly call out X as being where tech discourse is and then said you would use LinkedIn and substack but not X?

It is news to me that substack is where tech discourse. I just thought it was where paywalled articles go to die and give me endless pop-ups.
December 19, 2025 at 1:38 AM
No ethical consumption under capitalism and all that. And I totally get “Wordpress is a barrier”. But I didn’t quite get the “and so I must use substack” next step.

I feel like there’s missing info in the article? Does substack lead to more community engagement you can’t get elsewhere? Etc.
December 19, 2025 at 1:25 AM
Maybe I'm missing something, but I don't actually see a 'why'. Specifically, why substack is okay? I see a list of reasons it is bad and I see a an "I'm going to use it anyway to engage with the community and lower the writing barrier", but no real justification for why supporting substack is okay?
December 19, 2025 at 1:11 AM
Dinner for 14.5 hours or switching between 12- and 24-hour clocks? 😂
December 13, 2025 at 7:27 PM
Apple’s woven cables are my cat’s favorite to chew on.
December 12, 2025 at 8:01 PM
Except using these licenses has been a disaster for many companies as they are too vague to be "open". See HashiCorp with Terraform vs OpenTofu and the huge "binding FAQ" to clarify the vague BUSL. Even Elasticsearch walked their SSPL use back and offer AGPL now. CockroachDB abandoned BUSL.
December 5, 2025 at 12:17 PM
I think the internet would judge me harshly if I replaced Gateron Brown with Cherry MX Brown, even though I much prefer the Cherry as they are lighter weight.
December 4, 2025 at 8:04 PM
This inspired me to test a bunch of switches, including TTC Bluish White and Kailh Box Winter, and honestly not feeling enough of a difference from the Gateron Browns to justify getting new switches.

I need something maybe even lighter. Debating going linear with Kailh Box Silent Pink?
December 4, 2025 at 8:03 PM
I choose Sheb Wooley (Wilhelm scream guy)
December 1, 2025 at 4:29 PM
It unfortunately sets some flags as well so it’s not always possible to use. (e.g., Go’s assembler previously would always use XOR instead of MOV for zeroizing, which made it impossible to do certain bignum optimizations without ugly workarounds.)
December 1, 2025 at 4:26 PM