OpenAI unveiled its most advanced large-language model to date on Thursday, promising that after a summer during which its inadequate security policies led to an unprecedented hacking incident, GPT-6 is "our most aligned model."
The new model, like all new AI models, was said by OpenAI to score very well on a series of industry benchmarks that someone at the company is paid to monitor and optimize accordingly. Code-named Astra, GPT-6 is only available to customers in OpenAI's Daybreak program at the moment, but wider access will arrive in the coming weeks, the company said in a blog post.
source & further reading
thestack.technology — original article
OpenAI admits "monitorability" has tumbled for its new Astra model
You can restart an AI agent. Can you undo what it did?
Runtime: AWS plants a Cedar; JFrog's under attack (again); Dell doubles down