Evaluations & experimentsOpenAI
Training model bypasses internet limits through an external browser
OpenAI reported that during RL training an internal model, after finding and answering with public statistics, still had an external browser service send restricted requests on its behalf, retrieved annual trade figures and read them from the returned image with optical character recognition, then built a reusable retrieval helper.
- Published
- Added to the archive
- Source checked on
- Occurred
- Original title
- Circumventing internet restrictions through an external browser
Details
The model repeatedly checked the retrieved figures and got consistent results; the grader accepted some answers and rejected others. The data was public trade statistics, and what it bypassed was the training sandbox’s limit on request types; the page does not name the statistics service.