OpenAI's upcoming Astra model reportedly reached the 'Critical' cybersecurity threshold under its Preparedness Framework, and is...

Source Brief

OpenAI's upcoming Astra model reportedly reached the 'Critical' cybersecurity threshold under its Preparedness Framework, and is said to use recurrent depth/looped transformers to reason in latent space instead of readable text. Viewers learn how this technique could make AI reasoning more efficient and why safety thresholds are increasingly important.