سروش

ML researcher with a little bit of ADHD. Trying to escape from local minima

[سروش @ADHD_AI]

@damavand5555 درود بر شما. ایا منبع خوبی برای یادگیری رست و دم کردن قهوه سراغ دارید؟‌

[سروش @ADHD_AI]

🧠 What does a 3D Vision Model actually "see"? 👀

We cracked open Uni3D's transformer and visualized where the CLS token focuses its attention 🔬

Watch it classify a plant 🌱 — layer by layer 👇

[سروش @ADHD_AI]

@marwan_ptr @alzugarayign @makezur @XinKong_IC @AjdDavison I think you would like our work, in which we have used the token eviction idea on VGGT!
Check it out here:
https://t.co/n3oGpFGts1

[سروش @ADHD_AI]

@zhenjun_zhao @mhdjavanmardi Check the project page for code and more information:
https://t.co/n3oGpFH1hz

[سروش @ADHD_AI]

Our new paper. Code will be available soon.

[Zhenjun Zhao @zhenjun_zhao]

Evict3R: Training-Free Token Eviction for Memory-Bounded Streaming Visual Geometry Transformers

Soroush Mahdi, Fardin Ayar, Ehsan Javanmardi, Manabu Tsukada, @mhdjavanmardi

tl;dr: StreamVGGT+important token retention

https://t.co/r1LWvy6FT7

[سروش @ADHD_AI]

@breathlessrhino اینجا برا خاندان دریک رسما جلسه تراپی گذاشته :))

[سروش @ADHD_AI]

@Fa_ya_ha فیلم‌های Wong kar wai اگه وایبشون رو دوست داشته باشی خیلی خوبن

[سروش @ADHD_AI]

@Ali__bv72 Qv2ray

[سروش @ADHD_AI]

@handeredveyz اکثر اهنگای five finger death punch

[سروش @ADHD_AI]

درباره ارزیابی مدل‌های هوش مصنوعی و مقایسه‌شون با عملکرد انسان‌ها:

[François Chollet @fchollet]

Don't score AI using tests designed for humans. In particular because, with humans, the default assumption is that *they haven't already seen* the content you're giving them. With a LLM, the default assumption should be that, *if it's on the Internet, it's already been memorized*