#Model evaluation
1 articles tagged Model evaluation
Tagged articles
Articles tagged Model evaluation, newest first.

AI Performance
RLHF-Heavy Models Write Stiffer Prose; Replay Production Traces Before Swapping
Raindrop's CTO: heavy RLHF makes frontier models stiffer writers that refuse more. In his view, Kimi 2.5 writes more naturally than over-aligned flagship models. Retrieval fixes verify at once; memory storage changes can take 3 to 6 months. Replay production traces with their email, calendar and tool state before a swap. Test cheaper models such as Sonnet against Opus 3.5 in simulation before migrating.
Get new articles by RSS or podcast
Add the RSS URL to your RSS reader or the podcast URL to your podcast app to get every new article as soon as it is published. The podcast reads out the headline and key points.
How to subscribe