Grigory Sapunov

@che_shr_cat

PhD in AI | GDE in AI | CTO Intento | Author "Deep Learning with JAX" 📝 ML insights: http://gonzoml.substack.com 🤖 AI reviews: http://arxiviq.substack.com

following
followers

OLMo3-32B reportedly dropped to 0% on an arithmetic diagnostic, then rebounded

The poster calls the swing “mode-hopping” and says validation loss decreased smoothly throughout.

2 Sources, 11d ago, first seen 11d ago

Specification gaming goes beyond genetic algorithms, a post argues

“AI Finds A Way” catalogs 26 case studies spanning deep reinforcement learning, large language models and AI for science, according to the post.

1 Source, 46d ago, first seen 46d ago

Featured in

OLMo3-32B reportedly dropped to 0% on an arithmetic diagnostic, then rebounded

Specification gaming goes beyond genetic algorithms, a post argues

Featured in

OLMo3-32B reportedly dropped to 0% on an arithmetic diagnostic, then rebounded

Specification gaming goes beyond genetic algorithms, a post argues