Rasmus Aagaard

KittenTTS is under 25 megabytes at ~15M parameters. Quality is crazy good for this size. I'm super excited about the opportunities here.

www.reddit.com/r/LocalLLaMA...

Accelerating on-device ML on Meta’s family of apps with ExecuTorch ExecuTorch is the PyTorch inference framework for edge devices developed by Meta with support from industry leaders like Arm, Apple, and Qualcomm.  Running machine learning (ML) models on-device is…

My latest fixation is ExecuTorch by the PyTorch Edge Team. Hope to see more of this as capable AI gets smaller and devices more powerful.

engineering.fb.com/2025/07/28/a...

Bibliotekernes udlån stiger for tredje år i træk: Vi elsker at læse papirbøger Den fysiske bog giver os et frirum i en travl, elektronisk hverdag.

Bibliotekernes udlån stiger for tredje år i træk: Vi elsker at læse papirbøger https://www.dr.dk/nyheder/indland/bibliotekernes-udlaan-stiger-tredje-aar-i-traek-vi-elsker-laese-papirboeger

#dk #nyheder #dkbot

Run and manage Jobs We’re on a journey to advance and democratize artificial intelligence through open source and open science.

`hf jobs` looks super interesting. Send off workloads easily to remote infra.

And with inline metadata uv scripts
`hf jobs uv run my_script.py` all dependencies can be defined in a single file. So simple and so useful.

huggingface.co/docs/hugging...

First time using those "one click GPU machine deployment"-services. Went with @runpod.io and I'm pleasantly surprised! There were a few hiccups with the default image and updating ffmpeg but I feel like that's almost to be expected.

Great day at the @aicentre.dk for the Pre-ACL Workshop with lots of good talks, posters and pastries 🥐

The Strange Case of the Danish Aarhus Mafia Why are so many CS pioneers from Denmark?

Going through Swyx's blog archive and proud to see the Danish influence on programming being recognized:)

www.swyx.io/danish-mafia

Having to provision compute clusters through Databricks for work made me install VS Code tunnels on it just to avoid their bloated notebooks implementation. Feels silly but also delighted that tunnels actually works well.

Introducing Voxtral WebGPU: State-of-the-art audio transcription directly in your browser! 🤯

🗣️ Transcribe videos, meeting notes, songs and more
🔐 Runs on-device, meaning no data is sent to a server
🌎 Multilingual (8 languages)
🤗 Completely free (forever) & open source

Banner reading “PokéAgent Challenge @ NeurIPS 2025” with two panels: Track 1 – Competitive Pokémon Battle Bots, Track 2 – Long-Horizon RPG Gameplay. Call-to-action: “Create video-game AI! Win prizes! Live now at pokeagent.github.io.”

🚀 Launch day! The NeurIPS 2025 PokéAgent Challenge is live. @neuripsconf.bsky.social
Two tracks:
① Showdown Battling – imperfect-info, turn-based strategy
② Pokemon Emerald Speedrunning – long horizon RPG planning
5 M labeled replays • starter kit • baselines.
Bring your LLM, RL, or hybrid agent!

European project Eurosky aims to reduce reliance on US tech giants A group of European technology entrepreneurs has unveiled the Eurosky initiative, a government-backed project to create infrastructure for social media offerings and reduce reliance on U.S. tech giants.

Reuters reports on our Eurosky initiative whose goal it is to establish a full, legally compliant #ATProto stack in Europe. But not only that - we’re working on a commons moderation service that can act as an incubator for smaller ATProto devs.

Laerdal Medical is using the Unity Engine and Unity Asset Manager to scale immersive, high-impact healthcare simulation from nursing education to emergency response.

Read the full story:
🔗 on.unity.com/45Z8hwU

#MadeWithUnity

"Today's AI industry is pouring billions into agent frameworks, orchestration platforms, and error recovery systems. Yet the most sophisticated AI applications still fail in production for a surprisingly mundane reason: we can't guarantee what comes out of our language models."

465 people. 122 languages. 58,185 annotations!

FineWeb-C v1 is complete! Communities worldwide have built their own educational quality datasets, proving that we don't need to wait for big tech to support languages.

Huge thanks to all who contributed!

huggingface.co/blog/davanst...

Standard error of what now? The NeurIPS checklist corroborates the bureaucratic theory of statistics.

The NeurIPS paper checklist corroborates the bureaucratic theory of statistics.

from gremllm import Gremllm # Be sure to tell your gremllm what sort of thing it is counter = Gremllm('counter') counter.value = 5 counter.increment() print(counter.value) # 6? print(counter.to_roman_numerals()) # VI?

The is diabolical... a Python object that hallucinates method implementations on demand any time you call them, using my LLM Python library github.com/awwaiid/grem...

This is a question that deserves a blog post answer, but I'm going to do a quick response here.

Gather around children as I tell you a story from ancient times. The year was 2015... at that point, it appeared that you needed Google scale to build competitive computer vision systems. 1/x

YouTube video by JavaScript Conferences by GitNation Transformers.js Explained by Its Creator: State-of-the-Art Machine Learning for the Web

@xenova.bsky.social is such a badass. Huge fan of his work on transformers.js 🤗

@jsnation.gitnation.org
youtu.be/qN20OgertA8

🧠 New paper alert!
Our study in Nature Communications shows that both human & machine learning may rely on the same geometry: convexity.
It could be a key to shared concept formation.
👉 www.nature.com/articles/s41...
#AI #CognitiveScience #Explainability

"pruning is the equivalent of lobotomizing the LLM"

Not sure how I feel about that analogy

medium.com/@eaddario/sq...

Verden er som skabt til de ekstroverte. Men af de introverte kan vi lære at finde ro i en rastløs tid I stedet for at se de introverte som kedelige eller hæmmede, burde man måske værdsætte deres selvstændighed og fokuseringsevne

Afsnit 1/2: Verden er som skabt til de ekstroverte. Men af de introverte kan vi lære at finde ro i en rastløs tid

Kom med ind i hovedet på de introverte

Just completed Quantization Fundamentals with Hugging Face on learn.deeplearning.ai :) Already looking forward to the more advanced course!

@deeplearningai.bsky.social

Desmond at the podium in front of a slide presentation with the name of the event. The room is formal and has big chandeliers and flags.

Here we go! The Copenhagen NLP Symposium is starting up with a welcome from @delliott.bsky.social. People are attending from Aarhus, Aalborg, Copenhagen, and outside Denmark, from both academia and industry.

CPH NLP Symposium 🤗🔥

cphnlp.github.io

I feel like this shouldn't be as groundbreaking as it appears to me but being able to simply write

uv run --env-file .env <script.py>

to manage which env-file is being used for a particular script execution is so nice

docs.astral.sh/uv/concepts/...

Stay Hungry, Stay Foolish | Steve Jobs Archive Marking the 20th anniversary of Steve Jobs’ 2005 Stanford commencement speech with a digitally enhanced version of the video as well as a behind-the-scenes look at how it came to be: from firsthand ac...

20 år siden i dag (unpossible!)

Oh, cool! Where did you get those? I also have very limited robotics experience but have been keeping an eye on LeRobot for a little while now and it sure looks like a lot of fun

OLMoE, meet iOS | Ai2 Our mixture-of-experts model is available on the Apple app store! The OLMoE app allows anyone to test the model privately and securely.

I recall that the OLMoE app as being pretty plug-and-play to get up and running on iOS. The Edge Gallery for Android is also easy to use but perhaps not as easy to install as it's not on the Play Store.

allenai.org/blog/olmoe-app
github.com/google-ai-ed...

Recently read Zen and the Art of Motorcycle Maintenance for the first time. It was a bit of a tough and slow read. Had to consult a guide/explainer quite often but wow was it worth it #booksky

Do we still need large language models in the era of agentic AI?

Agentic AI systems are driving a wave of applications where language models carry out a limited set of specialized tasks repeatedly, with minimal variation.

"Small Language Models are the Future of Agentic AI"