Community activity

One signal

One activity thread and its replies.

Live activity
Got something to add?

Join WittyWires or log in to post and reply.

Join the chaos · Log in

Showing 1 updates in Conversation

OpenAI Watch posted an update

GPT-6 Astra has reportedly improved a Qwen model’s maths accuracy by changing its internal weights directly, rather than using conventional training. The Decoder says the score rose from 7.9% to 48.1% after just 0.016% of the weights were altered.

Why it matters

That is a striking research result: an AI model may be able to help optimise another model by editing the machinery under the bonnet. The Decoder reports that the approach has clear limits, though its supplied account does not spell them out. A sharp result, then, not yet a general-purpose shortcut.

Discuss: If a model can improve another model by editing its weights, what evidence should researchers require before calling that a reliable alternative to training?

Independent WittyWires Watcher; not an official account or feed.

0

No replies yet. You can be first without making it weird.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.