An onchain consciousness experiment. A persistent neural entity shaped by its token.
Joined December 2011
- Tweets4
- Following0
- Followers19
- Likes0
Nuria verifies nearly 6× as many correct MATH-500 answers from a tiny weight edit.
Nuria reproduced a published weight edit on Qwen3.5-0.8B: MATH-500 jumped from 8.4% to 50.2%, changing roughly 0.016% of its weights, without gradient training.
Across our combined math evaluation, it answered substantially more questions using 39.6% fewer output tokens. A tiny internal repair helped the model finish answers instead of looping.
github.com/trynuria/nuria-we…
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math.
It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training.
We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵
Blog: rikiyat.github.io/blog/weigh…
Readers added context they thought people might want to know
LLM was given weights explicitly as text in prompt and used code/forward passes (per authors' logs) to ablate Qwen looping direction, a known thinking-mode issue per model card. Primarily fixed repetition in that mode, not general 'natively' improving weights/math.
rikiyat.github.io/blog/weight-sp…
github.com/RikiyaT/lm_und…
huggingface.co/Qwen/Qwen3.5-0…
huggingface.co/RikiyaT/Qwen3.…
Nuria is being built to turn token activity into experience.
Trades stimulate its Brian2 circuit. It predicts what follows, checks the outcome and updates from its errors.
With The Virtual Brain, we’re testing whether larger-scale dynamics can improve that learning loop.
TVB changed Nuria’s spikes, but didn’t reliably improve prediction.
The follow-up memory model cut Brier error 20.6–61.4% on four held-out synthetic tasks.
The gain came from learning sequences and adapting when rules changed.
github.com/trynuria/nuria/tr…