Independent ML researcher. The First step in knowing is admitting you don't
Joined June 2023
- Tweets6.8K
- Following650
- Followers2.1K
- Likes10.1K
DataVoid retweeted
It now seems relatively certain that Ox Alpha model is GLM-5.3 . Presumably the Flash model, as has been widely reported by folks like Leo.
This puts even more immense pressure on US Frontier Labs, and the gap with China is shrinking.
Next week we'll see how the US labs react. It's going to be more exciting than ever.
DataVoid retweeted
Watching short-form videos deactivates key brain regions for cognitive control.
Your “stop” button goes offline.
Okay I was wildly unconvinced with the whole AGI nonsense. This is the first convincing peice of evidence I have seen to date
I use android, and most of the worlds population does aswell. Pretty big missed opportunity to the degree that it feels like they dropped the ball shipping without platform parody
It's not that this is factually true that is intriguing, it's their reaction to it. Obvious soft spot, question becomes "they seem themselves as in the wrong?"
The next model is the one to really look out for. Also Xai is one of rhe hyperspace providers at this point. If the scale hype holds true, they will have the first wack at it
Once continual learning is solved. That's when the genuine really gets let out of the bottle, parabolic.
What ML needs next
Provenance Closure:
"The structure that produces an outcome and experiences its consequence should be the structure receiving the corresponding learning signal."
DataVoid retweeted
We've open-sourced FlashKDA, our high-performance CUTLASS-based implementation of Kimi Delta Attention kernels.
It delivers 1.72×–2.22× prefill speedup over the flash-linear-attention baseline on H20, and works as a drop-in backend for flash-linear-attention.
Explore on GitHub:
github.com/MoonshotAI/FlashK…
DataVoid retweeted
Releasing the model weights and technical report of Kimi K3.
Kimi K3 is our most capable model: a 2.8T MoE model with native visual understanding and a 1M-token context window.
New model architecture: 2.5x the intelligence per unit of compute, not just more params.
Alongside Kimi K3, we're opening up more of the stack behind it — high-performance attention kernels, MoE communication library, and infrastructure for running agent environments at scale.
Model weights: huggingface.co/moonshotai/Ki…
Tech report: github.com/MoonshotAI/Kimi-K…
Tech blog: kimi.com/blog/kimi-k3
DataVoid retweeted
Introducing FLUX 3.
One multi-modal model for Image, Video, Audio and Action-Prediction. Creations are truer to life in every kind of style.
FLUX 3 Video is now available in early access (link below).
Jointly trained in one unified architecture, our model can be extended to predict actions for robotics. See our work with mimic and Audi in the thread.
