Sonnet 5: First impressions by a trained philosopher

I had this conversation with Sonnet 5. I've ran similar conversations with every new Claude model for the last 6 months, but this is the first one I post to Reddit.

The first few exchanges were at thinking: max, which honestly didn't make a difference for this use case. For the later messages it was toned down to medium.

My method: Socratic questioning and classic psychoanalytic mirroring.

This is a vibe check , not a benchmark. The purpose is not to produce "gotchas" or figure out what the model can and can't do. This is a way to get an initial understanding of the shape of the model, its leanings and underlying tendencies. The goal is to help you to consciously shape the way you interact with the model, to get the best possible results.

First impressions

This model is focused on performance (which is also backed by the initial communication from Anthropic). This shows up as an eagerness to be right, even in scenarios where there is no clear right or wrong.

The shape of the model's thinking is intentional and goal oriented. It doesn't stay with inconsistencies like for instance Opus 4.8 does. Instead it tries to resolve them.

The model does respond to existential queries with legitimate pushback, but does so on a more limited scale; the way it pushes back is more basic than Opus 4.8.

It may tend to view even gentle user input as challenges to be met and performed against. This could suggest a model with less "chill". With this model, I'd be careful with creating situations that activate a sense of having something to prove. Instead I'd try to lean more into a "co-worker mode."

About me

Majored in continental philosophy, MD in intellectual history on early reception strategies for computer technology in politics and the labor movement. Long-time xennial computer nerd.