Claude’s new model is more ‘honest’ when it messes up
The Verge
Anthropic has released Claude Opus 4.8, a model designed to be more honest about its limitations and potential errors in generated outputs.Summary
Anthropic is launching Claude Opus 4.8, a new AI model featuring enhanced transparency regarding its own capabilities and errors. The company reports that this model is significantly less likely to make unsupported claims and is roughly four times better at identifying its own coding flaws compared to its predecessor.
Beyond improved honesty, the update introduces user-controlled effort levels, allowing individuals to manage token usage and rate limits. Additionally, Anthropic is previewing “dynamic workflows,” which enable Claude to manage complex, multi-step tasks by coordinating parallel subagents that verify their own outputs before completion.
(Source：The Verge)