Rasmus Aagaard
@rasgaard.com
Industrial PhD Student @ Laerdal.com & DTU.dk, research in efficient machine learning, edge deployment and model compression
Open Source committee member @ DDSC.io
Open Source committee member @ DDSC.io
Proud to be working in an organization with a healthy attitude towards AI
This is one of our kick-ass designers telling how he used AI to rapidly prototype a new product that quickly had buy-in
This is one of our kick-ass designers telling how he used AI to rapidly prototype a new product that quickly had buy-in
Y Oslo - Pixel and prompts, Ola Helland
YouTube video by Y Oslo
www.youtube.com
November 7, 2025 at 11:18 AM
Proud to be working in an organization with a healthy attitude towards AI
This is one of our kick-ass designers telling how he used AI to rapidly prototype a new product that quickly had buy-in
This is one of our kick-ass designers telling how he used AI to rapidly prototype a new product that quickly had buy-in
It's not often that I want to share required reading material from a course but here we go.
My main takeaway is that being entrepreneurial is to be resilient by not planning too much, seek out opportunities and taking chances as they arise.
My main takeaway is that being entrepreneurial is to be resilient by not planning too much, seek out opportunities and taking chances as they arise.
(PDF) What Makes Entrepreneurs Entrepreneurial?
PDF | What are the characteristics, habits, and behaviors of the species entrepreneur? Is there such a thing as "entrepreneurial thinking"? Is there a... | Find, read and cite all the research you nee...
www.researchgate.net
November 6, 2025 at 12:33 PM
It's not often that I want to share required reading material from a course but here we go.
My main takeaway is that being entrepreneurial is to be resilient by not planning too much, seek out opportunities and taking chances as they arise.
My main takeaway is that being entrepreneurial is to be resilient by not planning too much, seek out opportunities and taking chances as they arise.
At Digital Tech Summit today and tomorrow :)
November 5, 2025 at 8:03 AM
At Digital Tech Summit today and tomorrow :)
Super nice to see the National Gallery of Denmark continuously being curious and embracing AI while acknowledging challenges
www.smk.dk/en/article/a...
www.smk.dk/en/article/a...
Kunstig intelligens (AI) på SMK | SMK – Statens Museum for Kunst
Hvordan forholder museet sig til kunstig intelligens?
www.smk.dk
October 28, 2025 at 9:23 AM
Super nice to see the National Gallery of Denmark continuously being curious and embracing AI while acknowledging challenges
www.smk.dk/en/article/a...
www.smk.dk/en/article/a...
What a wonderfully approachable blog post about a deeply technical topic
elanapearl.github.io/blog/2025/th...
elanapearl.github.io/blog/2025/th...
the bug that taught me more about PyTorch than years of using it
a loss plateau that looked like my mistake turned out to be a PyTorch bug. tracking it down meant peeling back every layer of abstraction, from optimizer internals to GPU kernels.
elanapearl.github.io
October 27, 2025 at 9:53 AM
What a wonderfully approachable blog post about a deeply technical topic
elanapearl.github.io/blog/2025/th...
elanapearl.github.io/blog/2025/th...
Reposted by Rasmus Aagaard
As someone who learnt so much by watching @shiffman.lol's coding videos in high school, I never imagined that one day my own library would feature on his channel! 🥹
If you're interested in learning more about 🤗 Transformers.js, I highly recommend checking it out!
👉 www.youtube.com/watch?v=KR61...
If you're interested in learning more about 🤗 Transformers.js, I highly recommend checking it out!
👉 www.youtube.com/watch?v=KR61...
October 26, 2025 at 7:30 PM
As someone who learnt so much by watching @shiffman.lol's coding videos in high school, I never imagined that one day my own library would feature on his channel! 🥹
If you're interested in learning more about 🤗 Transformers.js, I highly recommend checking it out!
👉 www.youtube.com/watch?v=KR61...
If you're interested in learning more about 🤗 Transformers.js, I highly recommend checking it out!
👉 www.youtube.com/watch?v=KR61...
"Så vi bliver simpelthen nødt til at holde op med at invitere ulven ind til de fora, hvor vi diskuterer, hvordan vi øger sikkerheden på vores hønsehus." - dhh, world.hey.com/david/tale-t...
Minister har udpeget Microsoft til udvalg med direkte indflydelse på uddannelsespolitikken | Radar
Microsofts danske education lead har med en ny udvalgspost fået direkte adgang til at påvirke udbuddet af uddannelser i Danmark. Hun modtager også et pænt honorar for hvervet.
ing.dk
October 26, 2025 at 3:50 PM
"Så vi bliver simpelthen nødt til at holde op med at invitere ulven ind til de fora, hvor vi diskuterer, hvordan vi øger sikkerheden på vores hønsehus." - dhh, world.hey.com/david/tale-t...
Really enjoyed reading kyutai.org/next/codec-e... and getting to know more about neural audio codecs
Neural audio codecs: how to get audio into LLMs
Why modeling audio is harder than text, and how to make it feasible with neural audio codecs.
kyutai.org
October 22, 2025 at 8:05 AM
Really enjoyed reading kyutai.org/next/codec-e... and getting to know more about neural audio codecs
Reposted by Rasmus Aagaard
correct
i’ve been saying this for a couple months. RL is driving towards specialization
my hunch is it’s temporary and something will shift again back towards generalization, but for now.. buckle up!
i’ve been saying this for a couple months. RL is driving towards specialization
my hunch is it’s temporary and something will shift again back towards generalization, but for now.. buckle up!
October 15, 2025 at 11:39 AM
correct
i’ve been saying this for a couple months. RL is driving towards specialization
my hunch is it’s temporary and something will shift again back towards generalization, but for now.. buckle up!
i’ve been saying this for a couple months. RL is driving towards specialization
my hunch is it’s temporary and something will shift again back towards generalization, but for now.. buckle up!
Testing out offline language model (SmolLM2-360M) along with in-browser database (pglite.dev with pgvector) for a completely local RAG system that runs on a phone.
Thanks @xenova.bsky.social for building so many demos and examples to build on top of! :)
Thanks @xenova.bsky.social for building so many demos and examples to build on top of! :)
October 15, 2025 at 11:44 AM
Testing out offline language model (SmolLM2-360M) along with in-browser database (pglite.dev with pgvector) for a completely local RAG system that runs on a phone.
Thanks @xenova.bsky.social for building so many demos and examples to build on top of! :)
Thanks @xenova.bsky.social for building so many demos and examples to build on top of! :)
Reposted by Rasmus Aagaard
AI-Gitto erstatter menneske-Gitte, og det skal vi vænne os til https://www.dr.dk/nyheder/indland/ai-gitto-erstatter-menneske-gitte-og-det-skal-vi-vaenne-os-til
#dk #nyheder #dkbot
#dk #nyheder #dkbot
AI-Gitto erstatter menneske-Gitte, og det skal vi vænne os til
På jobcenteret i Nyborg bliver man budt velkommen af Gitto. Hun er en AI-avatar og bare et eksempel på kommunernes øgede brug af kunstig intelligens.
www.dr.dk
October 10, 2025 at 12:54 PM
AI-Gitto erstatter menneske-Gitte, og det skal vi vænne os til https://www.dr.dk/nyheder/indland/ai-gitto-erstatter-menneske-gitte-og-det-skal-vi-vaenne-os-til
#dk #nyheder #dkbot
#dk #nyheder #dkbot
I’m experimenting a bit with writing thoughts about my PhD in public. Not sure what the format will or won’t be. Just thought that an earlier (and future!) version of myself would enjoy reading along :)
What has happened in the first week?
phd.rasgaard.com
October 8, 2025 at 5:38 PM
I’m experimenting a bit with writing thoughts about my PhD in public. Not sure what the format will or won’t be. Just thought that an earlier (and future!) version of myself would enjoy reading along :)
Reposted by Rasmus Aagaard
#EurIPS is getting closer and closer. There is now just 8 weeks until our the inaugural EurIPS conference and we cant wait to see you at the @ellis.eu UnConference, the EurIPS conference, 3 Affinity Workshops, and the 18 Workshops.
So if you haven't already, go get your ticket 🎟️🎟️🎟️
eurips.cc
So if you haven't already, go get your ticket 🎟️🎟️🎟️
eurips.cc
October 7, 2025 at 9:55 AM
wiki.c2.com?MakeItWorkMa...
1. Meet the minimum requirements for the business to call project a success. (Make it work.)
2. Add bells and whistles to make the program less prone to error and more feature rich. (Make it right.)
3. Find and eliminate waste in the process. (Make it fast.)
1. Meet the minimum requirements for the business to call project a success. (Make it work.)
2. Add bells and whistles to make the program less prone to error and more feature rich. (Make it right.)
3. Find and eliminate waste in the process. (Make it fast.)
measured improvement in server performance
wiki.c2.com
October 7, 2025 at 8:38 AM
wiki.c2.com?MakeItWorkMa...
1. Meet the minimum requirements for the business to call project a success. (Make it work.)
2. Add bells and whistles to make the program less prone to error and more feature rich. (Make it right.)
3. Find and eliminate waste in the process. (Make it fast.)
1. Meet the minimum requirements for the business to call project a success. (Make it work.)
2. Add bells and whistles to make the program less prone to error and more feature rich. (Make it right.)
3. Find and eliminate waste in the process. (Make it fast.)
October 7, 2025 at 8:19 AM
Reposted by Rasmus Aagaard
The EU is once again trying to push a dangerous legislative proposal that undermines encrypted communication and obliterates our rights to private conversations. www.eff.org/deeplinks/2...
Chat Control Is Back on the Menu in the EU. It Still Must Be Stopped
The European Union Council is once again debating its controversial message scanning proposal, aka “Chat Control,” that would lead to the scanning of private conversations of billions of people. Chat
www.eff.org
October 4, 2025 at 7:00 PM
The EU is once again trying to push a dangerous legislative proposal that undermines encrypted communication and obliterates our rights to private conversations. www.eff.org/deeplinks/2...
DHH's fireside chat at DTU was so great.
My main takeaway is that it's an amazing time to be optimistic and ambitious.
My main takeaway is that it's an amazing time to be optimistic and ambitious.
October 2, 2025 at 4:41 PM
DHH's fireside chat at DTU was so great.
My main takeaway is that it's an amazing time to be optimistic and ambitious.
My main takeaway is that it's an amazing time to be optimistic and ambitious.
First official day starting my new position as industrial PhD student at DTU.dk and Laerdal.com 🎉 So great to be back on campus.
Be sure to reach out if you're interested in model compression and edge deployment of deep neural networks 🔥🚀
Be sure to reach out if you're interested in model compression and edge deployment of deep neural networks 🔥🚀
October 1, 2025 at 2:29 PM
First official day starting my new position as industrial PhD student at DTU.dk and Laerdal.com 🎉 So great to be back on campus.
Be sure to reach out if you're interested in model compression and edge deployment of deep neural networks 🔥🚀
Be sure to reach out if you're interested in model compression and edge deployment of deep neural networks 🔥🚀
Reposted by Rasmus Aagaard
i wrote about atproto and why it matters
Open Social — overreacted
The protocol is the API.
overreacted.io
September 26, 2025 at 3:33 PM
i wrote about atproto and why it matters
Reposted by Rasmus Aagaard
something I’ve been thinking about is that all our content retrieval systems are biased towards recency rather than quality which makes it extremely hard to discover old good books, movies, blog posts, etc
September 19, 2025 at 6:24 PM
something I’ve been thinking about is that all our content retrieval systems are biased towards recency rather than quality which makes it extremely hard to discover old good books, movies, blog posts, etc
September 18, 2025 at 6:57 AM
Reposted by Rasmus Aagaard
🔁 If you are enjoying the feed, please like and share it with others for discoverability!
September 10, 2025 at 3:28 PM
🔁 If you are enjoying the feed, please like and share it with others for discoverability!
In anticipation of diving back into academia I finally read `You and Your Research` by Richard Hamming. I completely understand why it's so highly recommended
gwern.net/doc/science/...
gwern.net/doc/science/...
You And Your Research
Transcript of famous & widely-quoted 1986-03-07 lecture by Turing-Award mathematician Richard Hamming about how to do scientific research & development based on his life, antecedents of eminen...
gwern.net
September 8, 2025 at 1:35 PM
In anticipation of diving back into academia I finally read `You and Your Research` by Richard Hamming. I completely understand why it's so highly recommended
gwern.net/doc/science/...
gwern.net/doc/science/...
Reposted by Rasmus Aagaard
Syge børn, heroiske redningsfolk og nuttede kæledyr. En række Facebooksider deler systematisk AI-genereret indhold, der er skabt til at få brugere til at dele, like og kommentere. Og det virker. Men hvad er formålet? #smdk #AI #Meta www.tjekdet.dk/fup-og-spam/...
Er du hoppet i? AI-billeder på Facebook er skabt til at få dig til at reagere
“Forældreløs hele mit liv. I dag er jeg læge. Hvert eneste “tillykke”, jeg læser, betyder mere, end du kan forestille dig.”Sådan lyder billedteksten på et opslag, der har ramt hjertet hos over 15…
www.tjekdet.dk
September 3, 2025 at 9:16 AM
Syge børn, heroiske redningsfolk og nuttede kæledyr. En række Facebooksider deler systematisk AI-genereret indhold, der er skabt til at få brugere til at dele, like og kommentere. Og det virker. Men hvad er formålet? #smdk #AI #Meta www.tjekdet.dk/fup-og-spam/...
Haven't seen this before but it sounds super interesting. If your LLM application outputs a limited set of tokens why not prune away all the other unused output connections?
arxiv.org/pdf/2411.17713
arxiv.org/pdf/2411.17713
September 2, 2025 at 1:41 PM
Haven't seen this before but it sounds super interesting. If your LLM application outputs a limited set of tokens why not prune away all the other unused output connections?
arxiv.org/pdf/2411.17713
arxiv.org/pdf/2411.17713