Pinned post
New generation of ladies who are gossiping about distributed systems, FP, pragmatic architecture, chaos engineering and fast data analytics.
Lena Hall
Verified account This account is verified. Learn more@lenadroid
Senior Director, Developers & AI • Founder of Droid AI • Forbes Tech Council • LLMs • Data • AI/ML • Ex AWS, MSFT Research • Opinions are mine
5.5K Following 25K Followers 11K Posts
Latest posts 5
“hey, model, can you just make yourself better at this one thing?”
“sure, i just rearchitected my weights. all done”
fascinating
“sure, i just rearchitected my weights. all done”
fascinating
1/
What if an LLM could not just edit a model's weights, but re-architect it as it trains?
We tried just that. The resulting architectures exit the known territory of transformers: the coding agents spliced code directly into the network.
On nanochat, the loss fell below ordinary training after each new code layer was introduced. On algorithmic reasoning, accuracy rose from 55% to 84%. We even let DeepSeek-V4.1-Flash act as the agent modifying itself, raising its AIME and HMMT 2026 score from 64% to 71%.
To do all this, the agents built a set of tools we’ve never seen before, which we call mechanistic intervention!
Blog: doubleai.s.gy/llm-redesignView quoted post on X
back to SF next week!
looking forward to AI Summit, AI Engineer Code Summit, and a couple of private events
say hi if you’re there ✨
looking forward to AI Summit, AI Engineer Code Summit, and a couple of private events
say hi if you’re there ✨
Texas 🔥 💙
if you’re good at building software, designing, teaching, or solving one specific problem, you probably already have ideas that could be businesses
turning them into one means research, offer, website, finding people, and outreach. that stuff usually stays disconnected if you’re doing it in pieces
i made a new video where i take one of my existing projects and use @pazidotai to run the whole thing end to end
it gives you an operating officer that coordinates specialist agents under one business context. they handle the research, product work, and outreach together. you review and decide what moves forward
try it here: links.pazi.ai/lena
turning them into one means research, offer, website, finding people, and outreach. that stuff usually stays disconnected if you’re doing it in pieces
i made a new video where i take one of my existing projects and use @pazidotai to run the whole thing end to end
it gives you an operating officer that coordinates specialist agents under one business context. they handle the research, product work, and outreach together. you review and decide what moves forward
try it here: links.pazi.ai/lena
1:46