The New Claude 3.5 Sonnet is Lazy When Given Large Context

I've noticed a significant issue with Anthropic's latest Claude 3.5 Sonnet model: it becomes increasingly lazy as conversation context grows.

This is a consistent pattern where the model's willingness to complete tasks deteriorates with each addition to the conversation history.

The Pattern

The degradation follows a clear trajectory. Early in a conversation, Claude 3.5 Sonnet works as expected, generating complete responses and fulfilling tasks efficiently. But as context accumulates, it starts exhibiting increasingly "lazy" behaviors:

1. First, it begins breaking content into smaller chunks
2. Then it starts adding "[Continue with...]" placeholders
3. Later, it constantly asks for permission to continue
4. Finally, it reaches peak laziness - responding with nothing but:

   [Would you like me to proceed with the complete article content now?]

   (Without giving me the actual article or even parts of it.)

What I'm building

Delegate tasks. Get software.

Give Vroni a GitHub issue, bug report, spec, or rough idea. It reads the repo, plans the change, writes code, runs checks, and works toward a review-ready pull request.

Take a look at vroni.com