Amar Saini

Deep Learning Research

[Amar Saini @_Epoching_]

@luisgnet @abichungus 🤨📷📸

[Amar Saini @_Epoching_]

@kernel_trick Check out Llava-OneVision-2 and AutoGaze as well!

[Amar Saini @_Epoching_]

@luisgnet @yunbin_bae Fr, actually deleted mine last week

[Amar Saini @_Epoching_]

@maxbittker Big fan of this work, I eagerly await the benchmark update on every model release 😭 I hear Fable has some strong vision capabilities to play games at the “vision” level as opposed to underlying env access, any plans to test that out? (Or, is that what this has always been!)

[max @maxbittker]

Claude Fable 5.0 sets a new frontier score on runescape bench. y axis is log scale

[Amar Saini @_Epoching_]

@luisgnet @melodyshar Can confirm

[NVIDIA @nvidia]

This week at #CVPR2026, NVIDIA Research is presenting three papers across physical ai that offer groundbreaking solutions for training at scale across diverse applications:

→ GraspGen-X: the first foundation model for zero-shot grasping, trained on billions of simulated grasps

→ LCDrive: a model that replaces expensive text-based reasoning with compact latent representations

→ NitroGen: a generalized gameplay AI foundation model that harnesses NVIDIA Isaac GR00T to help train embodied agents

Learn more: https://t.co/H748YkAWS9

[Sihyun Yu @sihyun_yu]

Can MLLMs actually track what's happening in a video?
Introducing VSTAT 🎯, our new benchmark for visual state tracking.

The tasks are simple: count cups, read typed words, count page flips. Humans solve them easily. MLLMs don't.

https://t.co/dgqhqeVuSv

🧵 [1/11]

[Amar Saini @_Epoching_]

@skalskip92 I was curious on how well the language/prompt details can get captured, great example!

[Amar Saini @_Epoching_]

@skalskip92 Sweet experiments!

[Amar Saini @_Epoching_]

@luisgnet Aura

[Amar Saini @_Epoching_]

@luisgnet This album cover goes so hard

[Claude @claudeai]

Introducing Claude Design by Anthropic Labs: make prototypes, slides, and one-pagers by talking to Claude.

Powered by Claude Opus 4.7, our most capable vision model. Available in research preview on the Pro, Max, Team, and Enterprise plans, rolling out throughout the day.

[FFmpeg @FFmpeg]

@xiaonweb @AnthropicAI Because the patches appear to be written by humans

[Baifeng @baifeng_shi]

Humans can see in high-res, high-FPS in real-time. Why can't VLMs?

Introducing AutoGaze: ViTs/VLMs "gaze" only at key video regions! Up to 4-100x token savings, 19x speedup, and enables scaling to 4K-res 1K-frame videos.

📄 https://t.co/GhbWZwMAg7
🌐 https://t.co/mEJ991MAIR
🤗 https://t.co/FOfc2QRThi

(1/n)🧵

[Amar Saini @_Epoching_]

@SophieLWang Yea it’s fascinating, also inspiring to think many of these algorithms come from the 90s and such, super important foundation blocks!

[Sophie Wang @SophieLWang]

I made an interactive blog post about how JPEG image compression works: https://t.co/I3HU73R2ae

[Andrej Karpathy @karpathy]

Expectation: the age of the IDE is over
Reality: we’re going to need a bigger IDE
(imo).

It just looks very different because humans now move upwards and program at a higher level - the basic unit of interest is not one file but one agent. It’s still programming.

[Andrej Karpathy @karpathy]

sadly the agents do not want to loop forever. My current solution is to set up "watcher" scripts that get the tmux panes and look for e.g. "esc to interrupt", and send keys to whip if not present. Need an e.g.:
/fullauto you must continue your research!
(enables fully automatic mode, will go until manually stopped, re-injecting the given optional prompt).