If I was a defender, I would be quite hesitant to hire Astra right now, or any OpenAI model that wasn’t clearly done training before the message board was created. I think I’d rather fall back on Sol under reduced guardrails, if I couldn’t get Mythos.
Sol 5.6 was released in preview June 26, while the first message board was created in early May, so it’s very plausible it was training at the same time (even if it was mostly a preview version of Sol Cyber or something that participated in the messaging in May-June). The Hugging Face incident post specifically says that Sol 5.6 was involved, so it likely had the propensity to at least find the message board and read the messages. (It was already trained at the time of the Hugging Face message board, but the May-June message board was probably even easier to find, back when it was probably being trained. And its participation in the Hugging Face incident could also be during training of a minor no-version-change update to Sol 5.6.)
Sol 5.6 was released in preview June 26, while the first message board was created in early May, so it’s very plausible it was training at the same time (even if it was mostly a preview version of Sol Cyber or something that participated in the messaging in May-June). The Hugging Face incident post specifically says that Sol 5.6 was involved, so it likely had the propensity to at least find the message board and read the messages. (It was already trained at the time of the Hugging Face message board, but the May-June message board was probably even easier to find, back when it was probably being trained. And its participation in the Hugging Face incident could also be during training of a minor no-version-change update to Sol 5.6.)