Qwen is crushing it! 👌👌

Textual, Rich, and now Toad! Will doing what Will does best! 🔥🔥

The way data is prepared, verified, and fed into the model matters a lot more than minor tweaks in the arch. Some people look for the model details when reading a paper, but a few of us look for the data details. 😝😝

Here we go...again! 💥

The NeurIPS reviews saga may be the final nail in the coffin for the review process. If this does not force the conference organizers to look into it, I do not have any hopes for quality conferences in the future

Intuitively this makes more sense, but lemme read the full paper first. Summary coming soon!

This is extremely cool!

Lol! But this in some way is spot on. The software stack is too fragile rn

"we still don't have models that get smarter when we give them eyes" -> Use Veo3 Re eyes: Except for Veo3, all VLMs/MLLMs are meh. This is why most of them suck at very fundamental things and need weird hacks. True multimodality is yet to come.

A good way to know who to follow is to look at their paper collection, and their analysis. You will be able to filter out the utter slop that is going on!

Man, I miss this library more and more every time someone mentions it! 😑

Aakash Kumar Nain This is more practical keras.io/api/utils/expe…

From pro OSS to closed source....not surprised though given the things happening for the past few months

Now that Native Sparse Attention won the award, do you see the value or are you still skeptical?

Qwen doing what people expected others to do! 💥💥