And how can we fix it?
Useful post. Trapeze part was funny. First reasonable explanation I’ve seen for the weird DeepSeek word dumps, though not sure I believe it just yet. I will probably refer back to this if asked for takes about RLVR at ICML next week.
Useful post. Trapeze part was funny. First reasonable explanation I’ve seen for the weird DeepSeek word dumps, though not sure I believe it just yet. I will probably refer back to this if asked for takes about RLVR at ICML next week.