Lucas Beyer (bl16) (@giffmana) 's Twitter Profile
Lucas Beyer (bl16)

@giffmana

Researcher (Google DeepMind/Brain in Zürich, ex-RWTH Aachen), Gamer, Hacker, Belgian. Anon feedback: admonymous.co/giffmana

ID: 2236047510

linkhttp://lucasb.eyer.be calendar_today08-12-2013 13:31:09

13,13K Tweet

65,65K Takipçi

485 Takip Edilen

Lucas Beyer (bl16) (@giffmana) 's Twitter Profile Photo

Someone just sent me an anon note that I should check my scholar, and indeed, nice round number! Thanks anon person :) Tweeting this mainly to eternalize the screenshot for my future self.

Someone just sent me an anon note that I should check my scholar, and indeed, nice round number!

Thanks anon person :)

Tweeting this mainly to eternalize the screenshot for my future self.
Lucas Beyer (bl16) (@giffmana) 's Twitter Profile Photo

Speaking of sincos position embeddings, the main reason they were used in the original Transformer appears to be educated guesses; notice the language:

Speaking of sincos position embeddings, the main reason they were used in the original Transformer appears to be educated guesses; notice the language:
Lucas Beyer (bl16) (@giffmana) 's Twitter Profile Photo

The back view has the ears wrong and the boot on the wrong side. Clearly this thing does not have a world model, it’s just a stochastic painter, and we’re hitting a wall. Robin, maybe try harder!

Lucas Beyer (bl16) (@giffmana) 's Twitter Profile Photo

I wrote a short snippet (mini-blog) which has two facets: 1. Pixels have an area, keep that in mind. 2. Why did we use the esoteric `scipy.zoom` for resampling position embeddings in big_vision/vit.py? Enjoy: lb.eyer.be/s/pixels-have-…

I wrote a short snippet (mini-blog) which has two facets:

1. Pixels have an area, keep that in mind.
2. Why did we use the esoteric `scipy.zoom` for resampling position embeddings in big_vision/vit.py?

Enjoy: lb.eyer.be/s/pixels-have-…
Kevin Black (@kvablack) 's Twitter Profile Photo

These people copy + pasted the website I designed (including the body text), changed some stuff here and there, and then released it as their own work without attribution. The best part is they didn't even change the preview thumbnail so it still says "Octo". Incredible stuff!

Lucas Beyer (bl16) (@giffmana) 's Twitter Profile Photo

haha you guys are pretty funny, not gonna lie Jimmy Apples 🍎/acc That being said, give that poor plant a proper saucer, and don't go too hard on the shaky cam effect

haha you guys are pretty funny, not gonna lie <a href="/apples_jimmy/">Jimmy Apples 🍎/acc</a> 

That being said, give that poor plant a proper saucer, and don't go too hard on the shaky cam effect
Lucas Beyer (bl16) (@giffmana) 's Twitter Profile Photo

Llama 3.2 multimodal AND molmo on the same day?! Oh boy when will i find the time to read all that! Please not a third big announcement today, it’s already evening over here.

Lucas Beyer (bl16) (@giffmana) 's Twitter Profile Photo

Among all the yesterday releases and drama, we all kinda just missed this SigLIP + Gemma2-9B model from Alibaba. Seems to perform pretty well, and it even seems to use only public data, including some new data. How/why did we all miss it? huggingface.co/AIDC-AI/Ovis1.…

Among all the yesterday releases and drama, we all kinda just missed this SigLIP + Gemma2-9B model from Alibaba.

Seems to perform pretty well, and it even seems to use only public data, including some new data.

How/why did we all miss it?
huggingface.co/AIDC-AI/Ovis1.…