analiza aiqbench
The post reports that LLMs can interpret and edit a network's raw weights without gradient training, with one claimed result raising a small model's MATH-500 score from 7.9% to 48.1% by editing 0.016% of its weights. The headline results come from a paper's examples and a single blogger's laptop replication, and the frontier-model claims rest on a model that hasn't been independently benchmarked. For readers, the open question is whether weight-space editing generalizes beyond cherry-picked demos, and whether a small local model can match the reported behavior.
Qwen 3.8 27B on my laptop can interpret model weights, too
Yesterday’s hot paper was Sparks of Weight-Space Programming: LLMs can natively read & write neural network weights by Rikiya Takehi, Jayden Teoh, Alex Zhang, and Omar Khattab. The paper found that frontier models, in particular GPT-6 Astra, can both interpret and modify the raw weights of a machine learning model—literally a text description of the model containing a bunch of floating point numbers. I’ve been playing with one of the examples from the paper and I’m delighted to share that Qwen 3.8 27B running on my laptop can do this as well! Just by looking at the raw weights, GPT-6 Astra can tell what a network was trained to do. And without any environment feedback, it has no trouble writing, from scratch, small game-playing networks that outplay networks trained with PPO for 100 million steps. Likewise, asked to make Qwen3.5-0.8B better at math without gradients, Astra identifies an issue in its weights causing reasoning…
Extras afișat conform termenilor RSS ai publicației · 154 cuvinte · textul integral și figurile sunt pe site-ul publicației
Subiecte conexe
Disentangling MLP Neuron Weights in Vocabulary Space
arXiv · 6 aug. 2026
The limits of interpretability in multiple linear regression
arXiv · 16 iun. 2026
[AINews] LLMs can read & write neural network weights: "Weight Space Programming" cuts Autoresearch out of the RSI loop
Latent Space · 3 nov. 2026
REATS: LLM Reasoning-based Ensemble Learning for Adaptive Time Series Forecasting
arXiv · 21 oct. 2026