swyx on weight-space programming: we've reached the second section of "Exhalation"
swyx cites a research write-up from Latent.Space, saying that with weight-space programming, we have now reached the second section of Ted Chiang's short story "Exhalation." The cited content mentions that in an MIT study, GPT-6 Astra traced Qwen3.5-0.8B's loop of repeatedly consuming 32K tokens on math problems to two layers, and edited 0.016% of Qwen's weights, raising its MATH-500 score from 7.9% to 48.1% with sampling held fixed.
Why read this
AI score 66
The author borrows the metaphor of Ted Chiang's short story "Exhalation" to connect weight-space programming with weight editing in the cited research, offering a perspective for understanding modifications to model behavior.
Topics
View original postQwen3.5-0.8B often spent 32K tokens on a math problem without reaching an answer. In an MIT study, GPT-6 Astra traced the loops to two layers and edited 0.016% of Qwen’s weights. With sampling held fixed, its MATH-500 score rose from 7.9% to 48.1%. https://t.co/2QiKGksZpj
View original
Qwen3.5-0.8B often spent 32K tokens on a math problem without reaching an answer. In an MIT study, GPT-6 Astra traced the loops to two layers and edited 0.016% of Qwen’s weights. With sampling held fixed, its MATH-500 score rose from 7.9% to 48.1%. https://t.co/2QiKGksZpj
Body
with weight-space programming, we are now at paragraph 2 of exhalation https://t.co/lfNwVnT2Uh