Emre Kose

Computational Pathologist

[Emre Kose @emrecgty]

@sedielem Made me reminisce this paper:
https://t.co/eBJb1PMknB

Fig 3 is a beautiful demonstration of this kind of analysis imo, and the multiple experts for these distinct types of tasks in different denoising intervals totally made sense

[Emre Kose @emrecgty]

@gabriberton @iScienceLuvr DINOv2 does transfer quite well to pathology (in stark contrast to DINOv3). Current top models are almost all DINOv2 G/14.

And leveraging other sources is possible, extensively used by some groups that possess that data. Access to such data+slides is the main blocker usually.

[Daniel Han @danielhanchen]

We also released 1-bit Qwen3.8-27B quants which run on 8GB RAM.

They retain ~77% of accuracy compared to BF16.

We originally didn't want to release them but were shocked that they worked extremely well on our internal testing!

[Emre Kose @emrecgty]

@_xjdr Gives me this vibe

[Emre Kose @emrecgty]

@WitCelebs @ivanfioravanti It's actually an old and well known strategy, so I'd be surprised if it doesn't work.

https://t.co/OwqG1tiXiQ

[Emre Kose @emrecgty]

@hmyrbnl Bu konuyu sevmeyenleri ikna edecek, sevenleri mest edecek bir oyun tavsiyem var.

Elektronik devreler, kendi Assembly diliniz, ALU'larınız, CPU'nuz derken size sıfırdan komple (hardware+software) bir bilgisayarı yaptırıyor. Şiddetle tavsiye ederim.

https://t.co/tX5c8pL3Mc

[Emre Kose @emrecgty]

@miniapeur https://t.co/cnnwjnt9f7

[Emre Kose @emrecgty]

@emollick I think they really did fine-tune GPT4 on Slack messages in the end 😑

[Emre Kose @emrecgty]

@__tinygrad__ Making sure your library is smaller than the context length of the SoTA AI is a genius strategy tbh

[Emre Kose @emrecgty]

@jeremyphoward We all joke but some crypto bro actually did create a Q* coin TODAY and it already has ~100k volume. I swear this timeline doesn't make any sense.

[Emre Kose @emrecgty]

@mervenoyann I knew the fourth photo reminded me of a library
https://t.co/s5rjfuE35O

[JohnMark Taylor @johnmark_taylor]

NEW PAPER 🧵: Deep neural networks are complicated, but looking inside them ought to be simple. In this paper we introduce TorchLens, a package for extracting all activations and metadata from any #PyTorch model, and visualizing its structure, in as little as one line of code. 1/

[AI at Meta @AIatMeta]

Today we're releasing the Segment Anything Model (SAM) — a step toward the first foundation model for image segmentation.

SAM is capable of one-click segmentation of any object from any photo or video + zero-shot transfer to other segmentation tasks ➡️ https://t.co/qYUoePrWVi

[Jeremy Howard @jeremyphoward]

Our new course, "From Deep Learning Foundations to Stable Diffusion", is finally done after 8 months of work!!!

With >30 hours of video content (all free, no ads!), you'll learn how to create and train a Stable Diffusion model starting from pure Python 🧵
https://t.co/bK0PSIzFww

[AK @_akhaliq]

HuggingGPT: Solving AI Tasks with ChatGPT and its Friends in HuggingFace

abs: https://t.co/YWpFeBdNUq

[Lior Alexander @LiorOnAI]

This might be the most eventful week AI has ever seen:

Monday:
-Stanford Alpaca 7B

Tuesday:
-GPT4
-Anthropic releases Claude
-Google's PaLM API
-AdeptAI raises $350M
-Google adds GenAI to workspaces

Wednesday:
-Pytorch 2.0
-MidjourneyV5

Thursday:
-Microsoft 365 Copilot

[Aran Komatsuzaki @arankomatsuzaki]

Scaling up GANs for Text-to-Image Synthesis

- Orders of magnitude faster at inference time
- Can synthesize high-resolution images, for example, 16-megapixel pixels in 3.66 seconds.

proj: https://t.co/CfsJal8NK7
abs: https://t.co/DqxNrRIqNK

[Emre Kose @emrecgty]

@k_saifullaah 3407
https://t.co/cnnwjnt9f7

[Frank Willett @WillettNeuro]

Our new pre-print is out today! We demonstrate a brain-computer interface that turns speech-related neural activity into text, enabling a person with paralysis to communicate at 62 words per minute - 3.4 times faster than prior work.

https://t.co/jK1rdMT5Zl