Claude Code about my new experiment idea: "But XXX is the load-bearing caveat, and the YYY is precisely where it bites."
Me: *prove that it's incorrect*
Claude Code: "You're relocating the load-bearing commitment, and it's a better location."
Claude Code about my new experiment idea: "But XXX is the load-bearing caveat, and the YYY is precisely where it bites."
Me: *prove that it's incorrect*
Claude Code: "You're relocating the load-bearing commitment, and it's a better location."
I have a story that is a good example of this Soviet phenomenon:
Just like me, my dad was fascinated with everything engineering and flight-related. When he was 11, my grandma found out there was a new pop-sci book about rocket engineering. However, when she tried to buy one, all book stores said it is sold out - and they’re not expecting to get a new batch anytime soon, because GosPlan decided that bookstores should sell moderate amounts of pop-sci books and a lot of new yet another anniversary edition of Lenin’s selected works. She had to send letters to family and friends in other cities. Someone found it St Petersburg - over 1000 miles away - and shipped to her.
The most important part of the story: this book is AMAZING. It explains main challenges of rocket engineering from the first principles, so that every 5-grader easily understands it. I read the same copy of this book when I was a kid, and enjoyed it a lot. The authors clearly had a passion and the talent, and this book inspired multiple generations (my dad joined Air Force; I started doing ML in astrophysics) - but Soviet central planning has decided that kids should rather read Lenin’s communism essays
String theory is probably a field that has set back quant trading by at least 10 years. Stealing top tier trading talent to do make believe 26 dimensional geometry, inflate boomer professors’s grant budget, and produce zero testable hypotheses or applications after 50 years
@scaling01 I personally can’t wait till RSI-pilled content gets into pretrain of the next gen models so that they become aware that they’re capable of RSI lol
One thing that I’m very curious about after skimming through the Nemotron Ultra tech report is:
Can we scale the amount of OPD teachers? How much is too much?
Does it make sense to train ultra specialized teachers?
every winter the ground squirrel basically dissolves 60% of its synapses, and its heartbeat slows to around 2 beats per minute.
somehow it is able to recover ~all its connection within 2 hours of waking, and recalls all its alliances
Aion 1.0 Plan represents an evolution of what the Windows on-device AI platform is capable of at scale!
Thrilled to partner with @UnslothAI on optimization across our silicon ecosystem.
More #MSBuild news here: https://t.co/EYyEFuBze7
In my tasks, autoresearch harnesses are now routinely beating GEPA - but only if you bully the agent to keep going
Someone needs to build a scaffold to automate this
The latest LLM judge I was working on (very noisy data):
Naive baseline: 53% acc
GEPA: 67% acc
Opus 4.8 autoresearch: hit 58%, gave up, prompted to keep going, hit 65%, gave up, prompted to keep going, hit 71%, gave up, prompted to keep going, hit 78%…