Astra can read and write weights to improve Qwen

Re:Astra can read and write weights to improve Qwen by augment_me on Monday November 02, 2026 @02:41PM (#50342516)

The math example on Qwen was based on 50 batched iterations, so the model was able to evaluate it's changes on downstream tasks. And the fix was essentially to remove a direction in activation space that causes looping instead of an exit, this was already known to be a thing(all of the jailbreak/unleashed models do this) following Arditi et al. (Refusal in language models is mediated by a single direction, 2024)

So it's not able to insert new knowledge, but it's able to adjust the routing/looping/weighing of existing one.

As I understand, it can replace post-training like LoRA/GRPO, but if it'd cheaper the authors don't go into

[ Parent | on HN ]

Re:Astra can read and write weights to improve Qwen by giancarlostoro on Monday November 02, 2026 @03:50PM (#50343516)

> but if it'd cheaper the authors don't go into

Merging a model and a LoRa is insanely cheap, so assuming it didn't use any GPUs to do inspection etc, and just looked at a file on a FS, then I would be willing to say... It's cheaper than training a LoRa if that's the case, assuming Astra didn't take several weeks?

[ Parent | on HN ]

Re:Astra can read and write weights to improve Qwen by BoorishBears on Monday November 02, 2026 @09:26PM (#50347208)

The examples are presented like the findings build on each other, but they don't: Astra couldn't do anything with just numerical weights once they moved to Qwen.

[ Parent | on HN ]

Re:Astra can read and write weights to improve Qwen by epihelix on Monday November 02, 2026 @09:43PM (#50347338)

Yes, the key experiment was when they asked if Astra could improve the maths performance of Gemma 3 270M, a model which rarely looped, and it could not.

So, this is certainly very interesting (and very impressive regardless), but it's not nearly at the we-can-improve-everything-and-anything stage, even for very small LLMs.

[ Parent | on HN ]