Astra can read and write weights to improve Qwen

Re:Astra can read and write weights to improve Qwen by Nevermark on Monday November 02, 2026 @10:06PM (#50347515)

Next token prediction, if it requires solving math problems to get the next token is not "just" a next token predictor, its a machine that learned to understand math (as far as it was trained).

I am amazed people still don't know the difference between the shape of data (token prediction) and the actual problem (learning whatever is required to be able to predict the next token).

The former does not limit the sophistication of the latter in any way.

[ Parent | on HN ]

Re:Astra can read and write weights to improve Qwen by jgilias on Tuesday November 03, 2026 @12:17AM (#50348377)

I regret using that phrase. That was not “a talking point” I intended at all. But look how many HNers got nerd-sniped into talking about it instead of focusing on the actual point I was making. Oh well.

[ Parent | on HN ]

Re:Astra can read and write weights to improve Qwen by Nevermark on Tuesday November 03, 2026 @12:27AM (#50348439)

I did get that point! :)

(And you provided a valuable service/example for everyone who just thought the same thing! That phrase just keeps showing up. Whack! Whack! /h)

Unexpected free feature that plays right into recursive self-enhancement. What happens when they are trained to scale this ability up?

This could hit a wall, or be a step change in ... the rate of step changes.

[ Parent | on HN ]