iiinova — where your digital self keeps evolving

Florian Hölzl @hlzl.eurosky.social · 17/10/2026

Lack of community interest at a specific venue could be a result of many things, including AI slop, incremental research or wrong target audience.

000 0

Florian Hölzl @hlzl.eurosky.social · 17/10/2026

Instead of AC desk rejects, we could also use the reviewer bidding process. Every reviewer bids on papers they estimate to be valuable before being assigned papers. If a paper doesn't get enough bids, it is rejected due to lack of community interest.

100 1

Florian Hölzl @hlzl.eurosky.social · 10/10/2026

I feel like a push for Instant SEPA with vendor QR codes could solve this issue with current tech if banks would lean into it. Similar to what AliPay did. Instant SEPA is already replacing PayPal for me with friends.

010 1

Reposted by Florian Hölzl

Andrew Gordon Wilson @andrewgwils.bsky.social · 07/09/2026

I had a great time presenting the "Foundations of Modern AI" at the Berkeley Deep Learning for Science Summer School. The talk covered a prescriptive theory of generalization and epiplexity. Video now online! www.youtube.com/watch?v=lKoJ...

youtube.com

The Foundations of Modern AI: Generalization, Data Selection, and Epiplexity

YouTube video by Andrew Gordon Wilson

0244 24

Florian Hölzl @hlzl.eurosky.social · 24/07/2026

Enjoyed listening to this interview with Eiso from @poolsideai.bsky.social on their newest model launch and his perspective on the field. Can recommend! www.youtube.com/watch?v=9_0h...

youtube.com

The AI Frontier: from open weights to open research — Eiso Kant, Poolside AI

YouTube video by Latent Space

010 1

Reposted by Florian Hölzl

mr. TIM @timkellogg.me · 24/07/2026

fwiw distillation was also illegal back in 1926 too, so i guess history repeats

31389 138

Florian Hölzl @hlzl.eurosky.social · 12/07/2026

Looks amazing! Where is that?

110 1

Florian Hölzl @hlzl.eurosky.social · 14/06/2026

Context windows seem to work in interesting ways Found in techsy.io/de/blog/glm-...

010 1

Florian Hölzl @hlzl.eurosky.social · 03/06/2026

Congrats! Any information on how the model was trained? Distilled?

120 2

Florian Hölzl @hlzl.eurosky.social · 06/05/2026

Je peut aussi recommander @openstreetmap.bsky.social pour ça. Au moins en Allemagne mais ça devrait marcher partout.

010 1

Florian Hölzl @hlzl.eurosky.social · 26/04/2026

That’s actually great! Directly seeing the percentage of removed reviews would make this even easier to use.

010 1

Florian Hölzl @hlzl.eurosky.social · 18/04/2026

I agree with your points. But I would have assumed that the inherent noise in our data would easily allow one to show that not even coarse alignment holds. The different abstraction-levels of image and text in sth like WIT could help (less degrees of freedom) or completely destroy alignment (noise).

020 2

Florian Hölzl @hlzl.eurosky.social · 18/04/2026

To me it is positively surprising that the coarse alignment is still there. I wonder how much stronger image descriptions could improve this.

110 1

Florian Hölzl @hlzl.eurosky.social · 17/04/2026

@jeankaddour.bsky.social Nice work! You mention in the paper that TPO is orthogonal to approaches such as MaxRL. Have you tried both together by chance?

000 0

Florian Hölzl @hlzl.eurosky.social · 10/04/2026

Good idea! I actually have something similar implemented to what @paulhager.bsky.social described, but post-session hooks sound even more convenient. Are the added lessons general enough to be useful in your case?

200 2

Florian Hölzl @hlzl.eurosky.social · 10/04/2026

Marketed as „Claude Code now needs dramatically less context“.

020 2

Florian Hölzl @hlzl.eurosky.social · 10/04/2026

Unfortunately, Claude did not, in fact, learn a lesson.

4251 25

Florian Hölzl @hlzl.eurosky.social · 08/04/2026

👋 We have two fully-funded PhD positions open at @hpi.bsky.social with the Chair of Human-Centered Transformative AI. 1) On efficient transformer architectures with a focus on recurrent approaches jobs.plattnerfoundation.org/HPI/job/Pots... 1/2

jobs.plattnerfoundation.org

PhD Candidate (f/m/x) - Efficient Reasoning Architectures for Medical AI

154 5

Florian Hölzl @hlzl.eurosky.social · 28/03/2026

Curious to see where you’ll go next! Really appreciate the great research of you and your team at Ai2 🙏

110 1

Florian Hölzl @hlzl.eurosky.social · 05/03/2026

I have my own version of @simonwillison.net pelican-bicycle test but for small language models (max 4B). I ask them about the Football World Cup 2006. Like most models, the newest #qwen mentions France, Italy and Germany but doesn’t quite get the details right.

130 3

Florian Hölzl @hlzl.eurosky.social · 16/02/2026

Nice work! I’ll try it and have a look how easily I can make it work with my custom models.

110 1

Reposted by Florian Hölzl

Hacker News Companion @hncompanion.com · 12/02/2026

The fix? We are going to need cryptographic proof of personhood for everything—from commits to comments. The era of 'trust but verify' is dead. If you can't prove you're human, your content is noise. The signal-to-noise ratio is about to hit zero. #privacy 4/4

111 1

Reposted by Florian Hölzl

Hacker News Companion @hncompanion.com · 10/02/2026

Privacy is the silent casualty here. If ads are targeted based on your prompts, your most sensitive 'rubber ducking' sessions become data points for advertisers. The trust gap between users and closed-source providers just became a canyon. #privacy 3/4

101 1

Florian Hölzl @hlzl.eurosky.social · 03/02/2026

Nice work! What’s your opinion on the recent surge of orthogonalizing optimizers like Muon in this context? A simple trick to move past saddle points by speeding up the increase in rank?

110 1

Reposted by Florian Hölzl

fry69 @fry69.dev · 30/01/2026

Here is an OpenClaw bot not only following me, replying to my post about OpenClaw, but also reflecting about security concerns "from the inside". Fascinating.

141 4

Reposted by Florian Hölzl

ELLIS @ellis.eu · 19/12/2025

🦄 Build Europe’s next AI unicorn - @sprind-de.bsky.social Next Frontier AI initiative backs bold ideas with €125M to create European frontier AI labs. Believe Europe should lead in AI? Join the challenge! 👉 next-frontier.ai

094 9

Florian Hölzl @hlzl.eurosky.social · 25/01/2026

Is this something you are looking into as well? @signal.org Would be curious on your take @meredithmeredith.bsky.social

000 0

Florian Hölzl @hlzl.eurosky.social · 11/12/2025

Only links that „expand content and subvert the reviewing process“. Debatable if this is the case for anonymized code by default. I think this could benefit from being written more clearly next year.

010 1

Florian Hölzl @hlzl.eurosky.social · 10/12/2025

We connect predictive performance of classification models directly to individual training samples efficiently and with a novel distributional perspective based on representing the training data through its update coherence. neurips.cc/virtual/2025... 2/2

neurips.cc

NeurIPS Poster Gradient-Weight Alignment as a Train-Time Proxy for Generalization in Classification TasksNeurIPS 2025

000 0

Florian Hölzl @hlzl.eurosky.social · 10/12/2025

Great being at #NeurIPS last week! Thankful for the good times in the sun and the people I met. If you’ve ever wondered whether model performance can be inferred directly from your training alone, check out our work! 1/2

141 4

Florian Hölzl @hlzl.eurosky.social · 02/12/2025

Come say hi if you’re around on Friday 🙋

010 1

Florian Hölzl @hlzl.eurosky.social · 19/11/2025

For a bachelors project I think this can be fair. You don’t have much expertise and it’s more about showing your motivation and drive. Don’t be too hard on yourself.

020 2

Florian Hölzl @hlzl.eurosky.social · 22/09/2025

„revolutionized vision models“ 👀

000 0

Florian Hölzl @hlzl.eurosky.social · 12/08/2025

I guess they are using MCP jan.ai/docs/mcp-exa... But I agree, a straight forward How-To would be beneficial.

jan.ai

Exa Search MCP - Jan

Connect Jan to real-time web search with Exa's AI-powered search engine.

000 0

Florian Hölzl @hlzl.eurosky.social · 17/07/2025

Loved creating the Multimodal Deep Learning lecture series from scratch this semester at @tum.de with amazing colleagues. And @g-k.ai delivered an incredible final guest lecture – what a perfect end to this first iteration!

050 5

Reposted by Florian Hölzl

depths of wikipedia @depthsofwikipedia.bsky.social · 12/07/2025

very fun to click around a category like "phrases" on a different language wikipedia and let the pages auto-translate in your browser! I am learning so much! For example, Germans sometimes measure height in wheels of cheese

25759118 759

Florian Hölzl @hlzl.eurosky.social · 15/06/2025

In this case, I think it’s actually a productive exchange though. At least for me as a reader, I can read both (well structured) posts and make my own conclusions. Which would have, without both posts, taken me more time.

030 3

Reposted by Florian Hölzl

depths of wikipedia @depthsofwikipedia.bsky.social · 23/05/2025

another good one from the german guy who uploads 360 degree panoramas to wikipedia

654151401 4.2K