The decision to keep the AI's thought process hidden has raised concerns about transparency and interpretability, particularly among developers and programmers who work on making AI models safer. Critics argue that this approach centralizes responsibility for the language model with OpenAI, rather than democratizing it. The company's policy has been described as a step backwards, making its AI models increasingly opaque.
Key takeaways:
- OpenAI is reportedly threatening to ban users who try to get its AI model, code-named 'Strawberry', to reveal its reasoning process.
- Users have been receiving emails from OpenAI stating that their requests have been flagged for 'attempting to circumvent safeguards'.
- OpenAI argues that it needs to hide the chain-of-thought of its AI to maintain safety policies and to maintain a competitive advantage.
- The company's approach has been criticized for concentrating more responsibility for aligning the language model into the hands of OpenAI, rather than democratizing it, which poses a problem for programmers trying to make AI models safer.