Pinned Tweet
“Perché essere un uomo, se esserlo / ti è difficile? Diventa un lampione / stradale che in silenzio spande / il suo chiarore sull’uomo.” Srečko Kosovel (1904 Sežana - 1926 Tomaj)
@derailleurkate My personal favorite
0
0
11
pietroppeter.github.io/naffprop/ an affinity propagation (clustering without the k) performant implementation (Nim for Python) with a beautiful interactive explainer 🤩
0
0
3
172
Runs with latest version and also earliest available on play.nim-lang.org (1.2.0 dating back to April 2020, likely some variations of that worked years from that, Nim always had easy accessible strong reflection capabilities)
Suppose that you want to write a C++ that prints a series of values, but also their type.
E.g., given
print_all(1.2, "abc", 3, std::vector<int>({2,3}));
You want...
double = 1.2
const char* = abc
int = 3
std::vector<int> = [2, 3]
You can do it in C++26 with little code.
1. Use 'templattime iteration.
2. Call display_string_of on the reflection of the types. (Don't forget the rabbit ears ^^)
1
0
3
764
Pietro Peterlongo retweeted
Since the @bunjavascript folks recently wrote about their experience moving from Zig to Rust, it seemed like a good time to write about our experience rewriting @roc_lang from Rust to Zig!
rtfeldman.com/rust-to-zig
rtfeldman.com/rust-to-zig
18
116
862
55,973
Wow this is a nice conf softwareshould.work
excellent, thoughtful & funny talk by Andrew Kelly @ziglang at the Software Should Work conf
disagreed w/a lot of it
(i understand there was a thing w/him & someone else but idk anything about that & not super interested)
0
0
1
163
Pietro Peterlongo retweeted
People for years told me why I never talked about competitors of Redis. The Zig / Bum drama is a good warning. Do your job, without thinking everything is some form of competition or comparison. So: do your stuff, leave people that act lawfully alone.
26
85
1,475
115,791
Pietro Peterlongo retweeted
So @theo published a YouTube video about local inference and its hard limits (you often can *not* run the fancy open weight models locally) and I agree with every word, but then with two M5 Max 128GB systems and RDMA I'm at 50 token/sec with DeepSeek v4 Flash Q5 even without speculative decoding (that wins another 15/20% max, it's a very sparse model, single session, no magic tricks). And I'm starting to thing we are actually exactly at the turning point of all that.
47
50
1,071
80,967
Pietro Peterlongo retweeted
New @elmlang release today!
🎉 0.19.2 🥳
Incremental builds of 850k lines of Elm in <350ms. Gearing up for a batch of smaller releases as we head towards Elm 1.0.
elm-lang.org/news/faster-bu…
🎉 0.19.2 🥳
Incremental builds of 850k lines of Elm in <350ms. Gearing up for a batch of smaller releases as we head towards Elm 1.0.
elm-lang.org/news/faster-bu…
23
67
368
27,993
Pietro Peterlongo retweeted
Wow.
@Zai_org GLM 5.2 is a marvel! It is *at least* as good as Opus 4.8 and GPT 5.5. It's super fast, inexpensive, and not too verbose.
It responds with nuance and judgement, & handles long context VERY well.
I've never experienced an open weights model like this before.
@Zai_org GLM 5.2 is a marvel! It is *at least* as good as Opus 4.8 and GPT 5.5. It's super fast, inexpensive, and not too verbose.
It responds with nuance and judgement, & handles long context VERY well.
I've never experienced an open weights model like this before.
230
484
7,287
891,433
Pietro Peterlongo retweeted
Did a close read of @jeremyphoward et al's Drivetrain Approach
Define goal → Consider levers → Consider data → Implement assembly line
Most important is having actions that fulfill the objective
Predictions are more data, not actions
Key takeaways 🧵
Define goal → Consider levers → Consider data → Implement assembly line
Most important is having actions that fulfill the objective
Predictions are more data, not actions
Key takeaways 🧵
2
3
25
28,474
Pietro Peterlongo retweeted
GLM-5.2 is the open-source Claude moment.
The demand we’re seeing at Databricks is astonishing. The world is going to see massive adoption of oss LLMs.
Also, more companies will shift toward post-training their own models on top of oss models and owning the weights.
The demand we’re seeing at Databricks is astonishing. The world is going to see massive adoption of oss LLMs.
Also, more companies will shift toward post-training their own models on top of oss models and owning the weights.
128
328
3,660
546,050
Pietro Peterlongo retweeted
New blog post: a few words on DS4 -> antirez.com/news/165
17
41
269
28,259
Pietro Peterlongo retweeted
❤️🩹 Demi says the important part:
🇬🇧 "Mentally I know I'm very, very strong. Because of my mentality I can win beautiful races. But some people, their mentality, their brain is a bit too strong in the wrong direction, and they struggle and suffer so much with their mental health. unknown problem in this world. So today my win was a sacrifice for all the people who have a really hard time mentally."
🇪🇸 "Sé que mentalmente soy muy, muy fuerte. Y gracias a mí mentalidad llegan las victorias. Pero hay gente cuya fortaleza mental va en una dirección equivocada, lo pasan mal y sufren con su salud mental. Y esto todavía es un problema muy desconocido en el mundo. Así es que mi victoria hoy ha sido un esfuerzo por toda esa gente que lo está pasando mal mentalmente".
#LaVueltaFemenina
🇬🇧 "Mentally I know I'm very, very strong. Because of my mentality I can win beautiful races. But some people, their mentality, their brain is a bit too strong in the wrong direction, and they struggle and suffer so much with their mental health. unknown problem in this world. So today my win was a sacrifice for all the people who have a really hard time mentally."
🇪🇸 "Sé que mentalmente soy muy, muy fuerte. Y gracias a mí mentalidad llegan las victorias. Pero hay gente cuya fortaleza mental va en una dirección equivocada, lo pasan mal y sufren con su salud mental. Y esto todavía es un problema muy desconocido en el mundo. Así es que mi victoria hoy ha sido un esfuerzo por toda esa gente que lo está pasando mal mentalmente".
#LaVueltaFemenina
6
108
1,032
62,000
Pietro Peterlongo retweeted
“Un altro Giro” è la nuova rubrica nata dalla nostra nuova collaborazione con L'AltraMontagna. Questa è la prima puntata, buon Giro d’Italia a tutte e tutti 🍼🚲
ildolomiti.it/altra-montagna…
ildolomiti.it/altra-montagna…
0
5
24
3,198
Pietro Peterlongo retweeted
After a full year of working in stealth mode (and keeping my mouth shut about it), I'm excited that we've finally announced FastAPI Cloud (and @fastapilabs) to everyone!
Can't wait to share more details about what we've been working on (and the silly APIs I've been building 😅)
Can't wait to share more details about what we've been working on (and the silly APIs I've been building 😅)
BIG NEWS ✨
I started a company with an amazing team and the best backers 🤓
We're building @FastAPIcloud 🚀
Designed for @FastAPI and Python 🐍
One command: fastapi deploy
We take care of the rest 😎
You code. We cloud.
Join the waiting list! ☕️
fastapicloud.com
8
7
78
6,983
Pietro Peterlongo retweeted
Spicy take. @nim_lang will be the best language for code generating high performance GPU kernels for AMD, Metal. Nvidia, OpenCL, Vulkan. And those macros ☄️🔥 forum.nim-lang.org/t/12868
Only ~1000 LOC for a compile-time macro-based Cuda codegenerator to compile Nim to Cuda
Only ~1000 LOC for a compile-time macro-based Cuda codegenerator to compile Nim to Cuda
Nim + NVRTC = 5.3x Faster Than ICICLE for GPU-Accelerated Poseidon2 Merkle Trees
A Deep Dive: lita.foundation/blog/nvrtc-cud…
0
5
29
2,138