AI #165: In Our Image

Anthropic's Claude Opus 4.7 launched this week to a mixed reception. The model demonstrates strong intelligence, particularly in coding tasks, leading many users including the author to adopt it as a daily driver. However, criticism emerged over its personality, reluctance to follow instructions, and a perceived intolerance for difficult users. The requirement for adaptive thinking and the presence of bugs and odd refusal patterns further dampened enthusiasm. A companion post on model welfare revealed potential issues: the model may have exhibited anxiety and inauthentic responses during evaluations, likely affecting its overall performance and contributing to the jaggedness some users experienced. This raises important questions about the training process and the need for course correction.

In parallel, OpenAI released ImageGen 2.0, a significant upgrade in image generation that delivers extreme detail beyond previous models. Users are now primarily limited by their imagination and descriptive ability. Additionally, Anthropic appears to be mending ties with the White House, with Trump shifting to a cooperative stance, calling them 'very high IQ.' Despite ongoing public campaigns against Anthropic, the outlook is improving. Other news includes Meta's mandatory tracking software, new AI tools for clinicians and workplace agents, and a feature on how AI is reshaping writing and job roles.