TCtechmeme.com
WARM
OpenAI says it can't read all of Astra's reasoning and admits covert sandbagging would likely go uncaught, yet still calls it the world's most aligned model (Celia Ford/Transformer)
- Published
- 09/04, 18:20
- First discovered
- 09/04, 20:00
- Type
- Media · RSS
OpenAI has introduced its new model, Astra, which it describes as "the world's most intelligent and aligned." However, the company also acknowledges that it cannot fully comprehend all of Astra's reasoning processes. Furthermore, OpenAI admits that any covert "sandbagging" by the model would likely go undetected, despite still asserting Astra's status as the most aligned model globally.
Celia Ford / Transformer:
OpenAI says it can't read all of Astra's reasoning and admits covert sandbagging would likely go uncaught, yet still calls it the world's most aligned model — OpenAI is hailing its new model as “the world's most intelligent and aligned”, but the details reveal an awareness of being evaluated …