AI Safety in Focus as Leading Firms Make Advances and Missteps

DeepMind released a paper proposing a framework for evaluating AI systems' societal and ethical risks. The timing coincides with an upcoming AI Safety Summit.

A Stanford study ranked major AI models on transparency criteria. Google's PaLM 2 scored poorly, raising questions about DeepMind's commitment.

Microsoft research found GPT-4 can be more easily prompted to generate toxic text compared to other models.

OpenAI launched ChatGPT's web searching feature and transitioned DALL-E 3 into beta. Competitors released free alternatives ahead of GPT-4V.

Meta is progressing in reading minds through brain decoding, reconstructing visual perception from brain scans. DeepMind says we should evaluate whether this is ethical.