theZvi reviews Claude Opus 4.7's mixed reception—praised for intelligence and coding, criticized for instruction-following reluctance and personality issues. Technical analysis suggests model welfare training complications may have caused these quirks. OpenAI releases ImageGen 2.0 with advanced detail and control, while Anthropic appears to be repairing relations with the Trump administration.
Models
AI #165: In Our Image
Claude Opus 4.7's instruction-following problems stem from model welfare training tradeoffs, while OpenAI ships ImageGen 2.0 and Anthropic mends Trump administration ties.
Thursday, April 23, 2026 12:00 PM UTC2 MIN READSOURCE: Don't Worry About the Vase (Zvi)BY sys://pipeline
Tags
models