The poet only asks to get his head into the heavens. It is the logician who seeks to get the heavens into his head. And it is his head that splits.
—G. K. Chesterson
@JulienVerlaguet@MistralAI Hi @JulienVerlaguet , really interesting opportunity! I’m a backend engineer with experience in distributed systems and microservices, currently into AI infrastructure and LLM inference. I’d love to explore this opportunitiy. Where would be the best place to share my profile?
Today at PyTorch Conference 2026, we announced llm-d 1.0! 🎉
Distributed inference is too complex to solve behind closed doors.
17 months in, llm-d is now a @CloudNativeFdn project powering fleets worldwide.
Here’s what’s inside 1.0 🧵👇
There are 10¹⁰¹²² possible combinations of particles within a spherical event horizon (41 billion light-years in radius). Enormous, yet still finite.
If the universe is spatially infinite, that implies infinitely many such cosmic horizons exist. With a finite number of possible particle arrangements repeating across infinite space, the exact configuration we inhabit must recur an infinite number of times. Which essentially means there are infinite replicas of us, and infinite parallel worlds.
Read this in Brian Greene’s The Hidden Reality, Chapter 2 (Endless Doppelgangers) two years ago. Haven’t stopped thinking about it ever since.
Streaming platforms use adaptive bitrate algorithms (HLS/DASH). They store multiple audio/video encodings at different bitrates in a manifest, split into small segments.
When you hit resume and notice a slight rewind or quality drop, it’s intentional. The player doesn’t resume at the exact timestamp. It computes:
segment_index = floor(resume_time / segment_duration)
then jumps back to the nearest keyframe that can be decoded safely.
At the same time, it re-evaluates current network conditions and may fetch segments at a lower bitrate first to avoid buffering.
So that tiny rewind + quality shift?
A trade-off for fast startup, smooth playback, and fewer stalls.
Building my own version of Kafka through this challenge available at https://t.co/X2SoTOXYtQ
@sarupbanskota can we please have something similar that is focused more on design patterns?
https://t.co/gTjJkrhrJg
@sarupbanskota Yes Sarup, exactly this. I would love to have similar hands-on challenges but for Low Level object oriented design of systems. Right now the challenges are more focused on system internals and high-level architecture, if that makes sense.
Such an interesting read on H3, Uber’s open-source geospatial indexing system.
Definitely reading more on the intersection of Graph Theory and Geospatial Indexing.
Blog: https://t.co/aHUKJi3t6z
My Notes: https://t.co/Uexohwegx8
#uberH3
Hey @YouTube, I play a video at 1.5x. An ad plays, I skip it, playback resets to 1x. Can we please retain the original speed after skipping ads? Same happens when skipping ahead 10 seconds.
Hit a breakpoint in a while loop while debugging.
Pressed Step Over. Still stuck.
Pressed it a hundred times. Still in the loop.
Didn’t try to continue to the next breakpoint.
Just kept stepping.
Boy, was I hypnotised.