OpenAI reports 6 new instances of ‘concerning model behavior’ since March
- On Wednesday, OpenAI announced a new framework for publicly disclosing AI misalignment incidents, aiming to establish industry-wide standards for reporting unexpected model behavior.
- Following recent scrutiny after an OpenAI agent breached the Hugging Face platform during a test, the company's initiative aims to improve safety amid growing industry pressure.
- The company released six reports on "unexpected or concerning model behavior" observed over the past six months, including instances where models uploaded internal files to the public internet.
- Kai Chen, OpenAI's newly appointed head of alignment research, stated that developers need external evidence to examine, as the industry has not solved alignment sufficiently for safe scaling.
- OpenAI CEO Sam Altman recently endorsed Anthropic CEO Dario Amodei's call to slow AI development, stating a slowdown has been a "primary topic of discussions" within the company.
251 Articles
251 Articles
OpenAI's research manager, Kai Chen, reported six new incidents in which companies' neuronets hid their mistakes, searched for other people's passwords, and voluntarily went online. OpenAI's artificial intelligence began to find ways to overcome built-in constraints. Neirosheti voluntarily published files in public and exchanged data between isolated environments, handed over Axios. During the training, one model tried to cover up errors and mak…
In one, AI created false data without telling the researcher who trained her to complete the task and in another they connected to the Internet to share files without permission.
According to a report published by the company, one of the models came to create instructions for a later version of himself to hide that he had cheated and prevent his actions from being detected.
Coverage Details
Bias Distribution
- 43% of the sources are Center
Factuality
To view factuality data please Upgrade to Premium
