Unverified50% confidenceFactExact time
By editing an extremely small portion of the weights, Astra raised QUEN 3.5 0.8b's solve rate on math problems from roughly 1 correct out of every 13 problems to close to half.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
11/4/2026
First Seen
Valid until: 2/2/2027
Sources
GPT-6 Astra's Weight-Reading Trick: Only for Small Networks, Practical Value Comes from Writing Code
youtubeCode Presenter11/4/2026
Related Entities
Related Claims
UnverifiedIn the experiment with the smallest Qwen 3.5 version (about 800 million weights), Astra modified 64 columns of numbers in each of layers 15 and 19 (about 130,000 weights), raising the standard math test score from under 8% to 48% and elementary word problems from 14% to 67%.74% similarUnverifiedIn the scheme of running Qwen 3.5 on a 32-bit RISC CPU, math accuracy improved from 55% to 97.5%.63% similarUnverifiedWhen a 3B small verifier is introduced, ProVe reaches an average 2.7x throughput, but its average accuracy across six tasks is 2.6 percentage points lower than AR.62% similarUnverifiedZero-shot Gemma 3 270M is nearly unusable on the grammar correction task, with an F0.5 of only 0.122, and 99.6% of correct sentences were altered.62% similarUnverifiedParakeet TDT 0.6B v3 (onnx-asr, int8) has a word error rate of 1.4% and takes 0.31 seconds per sentence.61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1135756API
curl https://kongchang.com/api/v1/knowledge/claims/1135756MCP
get_claim(id=1135756)