OpenAI has decided not to release its latest iteration of ChatGPT, which it had labeled GTP 6.1 Astra. The company had safety concerns that researchers raised during internal testing.

The model was more capable than the company’s previous models in completing challenging tasks from end-to-end without human assistance, as well as writing.

But.

GPT-6.1 Astra regressed in two areas. Compared with its predecessor, GPT-6 Astra, the model performed poorly on tests measuring alignment, or how well the model adheres to what humans would like it to do. Specifically, GPT-6.1 Astra showed higher levels of deception: it wasn’t always honest about telling users of the actions it did or didn’t take.
Another issue was what OpenAI calls “scope authorization,” meaning that GPT-6.1 Astra would push ahead on a task without asking the user for permission, and would at times reach for external tools and services even if it might be unsafe.

This tale demonstrates pretty conclusively the criticality of humans having controlling involvement in AI sandbox construction, isolation, and enforcement and in testing. It also demonstrates the importance of corporate culture—i.e., corporate human managers—in creating and then enforcing the internal culture that supports and drives dedication to that human control of and involvement in development and testing.

It also proves that we humans are up to the task if we care enough about our responsibilities—closely akin to Madison’s and Franklin’s remarks that our republic requires a moral, virtuous people in order to survive.