Norman Casagrande
banner
adabstract.bsky.social
Norman Casagrande
@adabstract.bsky.social
ML, history, coding, space, science and everything in between. Personal profile. Research Eng @ DeepMind (Google). Views my own, etc.
Tired of replies? Join us on mastodon! ;)
June 6, 2025 at 4:08 AM
Correction: it *is* a federal republic. Apologies for the confusion.
June 4, 2025 at 5:11 AM
Switzerland is not a republic, but would it be long-lived enough to make its institutions interesting?
June 4, 2025 at 5:06 AM
100% agreed. I find it scary to see so many ppl trusting the output of those models. But I clearly have a bias having worked on their “innards”.
NotebookLM is the only one I find useful for research (with caveats) as it gives references to the sources you provide.
May 21, 2025 at 5:31 AM
It’s a classic!
May 5, 2025 at 7:01 PM
“I stand with 2000 years of darkness and bafflement and hunger behind me”. They don’t make those anymore 😅
youtu.be/yRujuE-GIY4
That Mitchell and Webb Look - Evil Vicar
YouTube video by mutant763
youtu.be
May 5, 2025 at 6:34 PM
Also because if you were smart but poor, the church was one of the few places that allowed you to get some education and a good chance for social mobility
May 5, 2025 at 6:28 PM
Reposted by Norman Casagrande
The training data distribution has a massive impact on learning. Imbalanced distributions (some individuals appearing more frequently) accelerate the plateau phase.
This suggests exciting new data scheduling strategies for training - we show that a simple warmup works well!
April 3, 2025 at 12:21 PM
What makes you think the rule of law would still apply? ;)
February 23, 2025 at 5:04 AM
Agreed. That’s a serious risk.
February 8, 2025 at 5:28 AM
A difficult but incredibly powerful read..
January 4, 2025 at 5:58 AM