Latent Space: 'The End of Finetuning' — Top 1% Still Fine-Tune More, Bottom 99% Can Skip It

Latent Space·medium signal

Latent Space's AINews argues finetuning is ending for most practitioners as long prompts and constitutions suffice, but the top 1% (Cursor, Cognition) have actually increased RLFT usage. The piece distinguishes between hobby/enterprise tiers: very long system prompts may be all most need, while frontier coding agents are investing more in specialized training. A key strategic question for any team deciding where to invest.