Unverified50% confidenceFactExact time
In a math repair experiment, a small open-source model scored only 7.9% on the Math 500 test; Astra raised the score to 48.1% by locating and subtracting a direction within two layers (about 130,000 weights, an edit ratio of 0.016%), and its instruction-following ability also improved roughly twofold.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
11/4/2026
First Seen
Valid until: 2/2/2027
Sources
Related Entities
Related Claims
UnverifiedWithout any training, Astra modified only 0.016% of the weights of a small model, raising its score on the Math 500 benchmark from 7.9% to 48.1%.80% similarUnverifiedIn the experiment, Astra modified the weights of the small Qwen model, raising the MATH500 score from 7.9% to 48.1%, while the share of answers hitting the trial-run limit fell from 93% to 33%.79% similarUnverifiedOne AI model's score on the MATH500 math test rose from 7.9% to 48.1%, an increase of about sixfold, stemming from modifying 0.016% of another model's weights.77% similarUnverifiedThe "Astra fixes Qwen's math ability" experiment used Python and an evaluation loop to find a steering vector that changed only a tiny fraction of the weights, raising the math benchmark score from single-digit percentage points to nearly half.74% similarUnverifiedOn math tasks, as the training answers become increasingly absurd, the proportion of misaligned responses soars from 4% to over 60%, while the model's general capabilities show almost no loss.73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1135779API
curl https://kongchang.com/api/v1/knowledge/claims/1135779MCP
get_claim(id=1135779)