I write about opportunities in science, space, and policy here: https://splittinginfinity.substack.com/
Loading...
NEW PAPER. Why do larger networks train better?
"Because they contain more candidate *sub*networks that can learn the task" → lottery tickets
This popular explanation uses an appealing but misleading metaphor🧵
We propose an intuitive alternative grounded in theory: escape dimensions
... and the cloud is just a decentralized storage and compute marketplace with exchange supported by cryptocurrency.
www.filecoin.io
Modern LLMs are incredibly good compression algorithms, which can shed light on why autonomous data science agents don't overfit as much as you might think. arxiv.org/abs/2606.11045
pruned inactive accounts, added new moots, was very difficult to squeeze down to just 150
go.bsky.app/6ncErBZ
What happens when we have an AI smart enough to solve any problem we can think of? That finishes every task on our checklist? We'd stop then right?
Conversely, Mythos will probably do great on arc-agi-3 given its ability to iterate on screenshots
At the end of the software singularity created by AI, I think the internet is going to look a lot like ATProto.
Posts and messages are just chunks of text you digitally signed. Platforms/apps/feeds/search are just an llm gathering this info and formatting it for you.
...
bsky.app/profile/hars...
Victor Taelin on Fable. "... my personal singularity moment"
Solves all his pet benchmark problems, achieved 22% speedup on the project he's working on in 2 hours of thinking (Opus got something similar but in 8 hours of work)
...
x.com/VictorTaelin...
