“DeepMind’s models will be reevaluated every time the compute power used to train the model increases six-fold, or is fine-tuned for three months. In the time between evaluations, it says it will design early warning evaluations.”
– Reed Albergotti
Google’s AI researchers are working to identify “critical capability levels” in future iterations of AI models so they can intervene. They will look for behaviours such as an AI model’s ability to manipulate human behaviour or generate malware.
Google DeepMind launches new framework to assess the dangers of AI models | SEMAFOR | May 17, 2024 | by Reed Albergotti