Tim (@daidailoh) header
Tim (@daidailoh) profile picture

Tim

@daidailoh

Dr. rer. nat. RWTH Aachen - founder https://teapaw.org/t.co/eSyuPfG0BT - LLMs, VLMs, CV - explaining science for @golem - Cosplay events at Sewcase e.V. - GameDev on full moon

 Aachen, Germany
Joined November 2015
  • Tweets790
  • Following461
  • Followers140
95Media
Photo by Tim (@daidailoh)Photo by Tim (@daidailoh)Photo by Tim (@daidailoh)Photo by Tim (@daidailoh)Photo by Tim (@daidailoh)Photo by Tim (@daidailoh)Photo by Tim (@daidailoh)Photo by Tim (@daidailoh)Photo by Tim (@daidailoh)Photo by Tim (@daidailoh)
Replying to @SchmidhuberAI @Pontifex
I did not expect this:
Image from Jürgen Schmidhuber (@SchmidhuberAI)
18
21
10
400
27,170
Never thought that visiting a conference can be healthy #ECCV2026
Image from Andrii Zadaianchuk 🇺🇦 (@ZadaianchukML)
1
13
4
96
16,665
BREAKING: Mathematician claims he already published the navier stokes proof in 1990, with OpenAI now stealing credit. In an interview, Jürgen S. further elaborated his accusations [...] #openAI #navierStokes
58
Happy to announce: I should now have the same level of access to Claude Mythos as the GOAT @karpathy ! Same goes ofc for Le Chaton Fat!
1
284
Two months ago, I vaguely posted a number: 0.9 FID, one-step, pixel space. Now it is 0.75, and can be even lower. Many wonder how. I thought it might end as a small FID prank: simple and deliberate. It started with one question: can FID be optimized directly, and what does it reveal? Introducing FD-loss.
Image from Jiawei Yang (@JiaweiYang118)
59
159
41
993
263,566
Sci-Hub is an evil website that pirated 85M+ research papers and made them freely available And now they've added AI to their database to make Sci-Bot. It answers your questions using latest, full-text articles. But DO NOT use it. We should all try to make billion-dollar academic publishers richer. I'm putting the link below so you know how to avoid it.
821
8,604
957
45,742
5,058,788
„Ich befürchte, dass die #Bundesregierung das Flickwerk der letzten Wochen fortsetzt. Nach ‚Tankrabatt‘ und ‚Klimageld‘ lädt der Bundeskanzler die Tarifpartner nun ein, sich auf #Einmalzahlungen einzulassen. Das ist kein Konzept, das ist Flickwerk.“ (tm) @morgenmagazin
281
570
342
1,510
Excited to announce our latest (submitted to) SIGBOVIK 2026 @sigbovik paper: "SchmidhubAI: Accurate Historical Paper Attribution". We built an AI system that, given any modern AI paper, automatically determines which of its ideas were already published by Jürgen Schmidhuber.
5
16
2
138
61,700
Ich weiß nicht, warum ich das getan habe, aber... Ich gebe euch: Angela Merzel. #Merz #Merkel
Image from Tim (@daidailoh)
173
Replying to @cb_doge
The actual cover should have credited the people whose research led to this AI boom @SchmidhuberAI
Image from Marko Njegomir (@njmarko)
27
69
13
654
239,505
i wonder if this is what it will be like when we get 10-trillion parameter LLMs
Image from Jack Morris (@jxmnop)
317
476
110
13,304
757,332
Very excited to announce that I'll be working on Meta's Llama now! (I just fiddled around with a local copy and wanted to join the latest "deep learning"-twitter trend to be cool like @giffmana et al. Still waiting for an offer from them ;))
1
1
223
Hey Zuck if you're reading this: Let me bring you natively 1D sequences to describe images - bring tokenised representations of images to the next level! Global tokens make much more sense than local ones :) arxiv.org/abs/2407.11913
Image from Tim (@daidailoh)
1
148
Happy to say: My baby, "Multidimensional Byte Pair Encoding: Shortened Sequences for Improved Visual Data Generation", got accepted into ICCV 🎊🍍
196
OpenAI should re-name their models more honestly - can't get much worse than how it is right now. Just name them "ChatGPT_final_2(copy)".
184
Extremly happy to say that I'll be presenting our paper, "Retargeting Visual Data with Deformation Fields", at ECCV in Milano - 2nd of October, Poster Session 4 at 16:30 :) github.com/DaiDaiLoh/image_d… Come say hi! (Please, I don't know anyone, would love to find people to get food😆)
GitHub - DaiDaiLoh/image_deformation_fields: Examplary code for images, for the paper "Retargeting...

GitHub - DaiDaiLoh/image_deformation_fields: Examplary code for images, for the paper "Retargeting...

Examplary code for images, for the paper "Retargeting Visual Data with Deformation Fields" (ECCV 2024) - DaiDaiLoh/image_deformation_fields

github.com
1
234
As a scholar, I'm deeply disappointed with the small count of "All You Need"-References in the titles :( #ECCV
Image from Tim (@daidailoh)
2
164
After being frustrated to death, I've added an example for an autoregressive, encoder only Transformer to my github repo of simple visual deep learning examples. It explains how to produce 12-by-12 pixel MNIST images with a transformer in below 200 lines github.com/DaiDaiLoh/Exempla…
ExemplaryNotebooks/MNIST_AutoregressiveImageTransformer.ipynb at main · DaiDaiLoh/ExemplaryNotebooks

ExemplaryNotebooks/MNIST_AutoregressiveImageTransformer.ipynb at main · DaiDaiLoh/ExemplaryNotebooks

Collection of simple examples in modified form, e.g. more stable GAN variants; All in ONE Notebook without many dependencies, no bells and whistles - just clean code examples - DaiDaiLoh/ExemplaryN...

github.com
3
165
Happy to share that #PointVOS has been accepted to #CVPR2024 🎉 It was a great collaboration with Sabarinath Mahadevan, Paul Voigtlaender, Bastian Leibe🥳 I will be in Seattle next week to present our poster😊 📜Paper: arxiv.org/pdf/2402.05917 🌐 Website: pointvos.github.io
1
7
21
1,472
Because I got nothing better to do, I started a github repo for super clean / super simple examples of modern deep learning networks, like GANs, VQ-VAEs, etc. - 99% of pytorch users should have everything installed already. Soon(tm): Diffusion, PointNet github.com/DaiDaiLoh/Exempla…
2
3
819
*added a WGAN example
83
Load more