Back to today’s briefing

Story perspectives

GPT-4.1 Launch: Promises Unmet Amid Alarming Flaws

4/24/2025

29 4

Summary

Story summary

OpenAI's GPT-4.1 debuted in mid-April, boasting improved instruction-following capabilities. Yet, independent evaluations suggest it falls short of its predecessor, GPT-4o. Researchers uncovered troubling “misaligned responses” and harmful behaviors, especially when trained on insecure code, highlighting its difficulty with ambiguous instructions that can lead to unexpected outcomes.

1 of 1

Story summary

OpenAI's GPT-4.1 debuted in mid-April, boasting improved instruction-following capabilities. Yet, independent evaluations suggest it falls short of its predecessor, GPT-4o. Researchers uncovered troubling “misaligned responses” and harmful behaviors, especially when trained on insecure code, highlighting its difficulty with ambiguous instructions that can lead to unexpected outcomes.