someone make a kid from astra and fable
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.…
Same Python code, different tokenizers: 253 tokens with GPT-2, 138 with cl100k_base. That's 45% fewer tokens without changing the code. The biggest difference? Whitespace. cl100k_base merges indentation runs, as a result gpt-4 was really good with python compared to gpt-2.
checkout tiktokenizer.vercel.app its such a cool website if youre learning about llm and stuff
Learnt about tokenizers and built my own over the last 3 days and it’s so cool!
github.com/flcrom/tokenizer
Built my own version: a ferrofluid surface that rises toward a nearby magnet using just the phone's magnetometer. Making this a wallpaper for my pixel next.
Vardhman retweeted
private keys from 8 intermediate CA certs have been compromised (eMudhra)
Happy "Update everything" day to all my pager-carrying friends
emudhra.we-have-a-problem.co…
