nyxkrage Profile Banner
Carsten Kragelund Profile
Carsten Kragelund

@nyxkrage

Followers
376
Following
1K
Media
54
Statuses
972

Always esoteric, always clever. Rock Whisper (Software Engineer). ❄️ Nix Devotee, DevOps is a sham by big tech 🐫 OCaml Evangelism Strike Force

Denmark
Joined October 2020
Don't wanna be here? Send us removal request.
@insaneisane
Isa Lansky
4 days
Opening emergency commissions! No theme/character restrictions, DM me if you're interested. Also open for short-term/long term gigs! Portfolio in the comments ⬇️ Sharing the post will be much appreciated, thank you 🙏
5
42
198
@snowclipsed
snow
2 days
new blogpost after a long time! in this series i will talk about how to solve reinforcement learning for long-horizon tasks, incrementally from the most straightforward approaches. (link in replies!) in part I of this series, we throw RL at the cube in its most direct,
21
40
337
@tenderizzation
tender
4 days
[ENG SUB] how it feels to use eager pytorch in 2025
24
55
406
@vikhyatk
vik
1 month
@AllanatrixQ it is locktober after all
0
2
10
@m_sirovatka
Matej Sirovatka
8 days
you can just create environments
@willccbb
will brown
8 days
creating an environment with verifiers is as simple as writing a load_environment function and filling out a pyproject.toml, both of which are initialized for you when you do `prime env init` environments are packages, and can be used with prime-rl, skyrl, tinker, and more :)
1
7
65
@davesnx
David Sancho
11 days
This is such a great idea that I implemented an ocaml port on my way back from Singapore It will be handy internally at ahrefs for all LLM structured outputs! https://t.co/TpGtVxpQ3S
@jschopplich
Johann Schopplich
17 days
JSON is token‑expensive for LLMs – just like @mattpocockuk frequently mentions. Meet TOON, the Token‑Oriented Object Notation. 💸 40–60% fewer tokens than JSON 📐 readable & tokenizer-aware Wrap your JSON with `encode` to save half the token cost: https://t.co/UoG9yHmgfg
4
4
26
@redtachyon
Ariel
13 days
It's only fine-tuning if it comes from the low-rank region of the parameter space. Otherwise it's just spicy retraining.
0
4
28
@SoupRS
Stefan
16 days
Nine years ago today, I interrupted the official RuneScape livestream with a profanity I regret screaming in the innocent ears of thousands.
73
123
2K
@_ueaj
ueaj
16 days
New blog post! This one is a purely theoretical one attempting identifying the central reason why LLMs suffer from mode collapse in RL and fail to generate novel or truly diverse outputs. It's actually a way more complicated problem than you think! Naively encouraging
30
74
674
@jdchawla29
jellybean ❄️
17 days
in another life i would love to get stuck in an elevator with you
0
5
48
@nyxkrage
Carsten Kragelund
17 days
Men are like let's see that tummy
@peach2k2
peachey 𔐓
17 days
Men are like let's see the personality
1
0
6
@schwarzposter_
neurotica
17 days
13
12
505
@redtachyon
Ariel
18 days
4
17
392
@humblemudgames
Humblemud
19 days
Last week, I covered 12 amazing, ✨new✨ games made for the #ZXSpectrum , a personal computer from the 80s. This week, I've got 12 MORE for you that might be even better.
1
3
10
@nyxkrage
Carsten Kragelund
20 days
The about:config flags to do this is 1. https://t.co/AutXm4yUQW.enabled to true 2. https://t.co/AutXm4yUQW.provider to any chat site that supports the ?q url parameter to start a new chat with a specified message 3. Then to use custom prompts/shortcuts set any of the
@MangoSweet78
Mango
20 days
Fun fact: You can use Firefox and hack in custom prompts to use w/ alongside the sidebar menu where you can hook up Open webUI and local LLM models. Don't give into the OAI-goonery.
0
1
2
@IlyasHairline
Ilya Sutskever's hairline
21 days
hensen chat is very special today
1
2
12
@johannes_hage
Johannes Hagemann
21 days
@Grad62304977 guy who is 19 years old: "I never really got 10+ year timelines"
20
8
276
@IlyasHairline
Ilya Sutskever's hairline
21 days
hensen chat reveals the truth
0
1
4
@vikhyatk
vik
1 month
@reflection_ai HF link?
3
4
115
@vikhyatk
vik
1 month
i'm hiring
28
23
363