We're back after one week away. Here's what we collected for weeks 2 and 3 -

Policy

Americas:

• Schumer’s AI game plan, by the numbers
• America's first law regulating AI bias in hiring takes effect

Europe:

• Founder and investor, Ian Hogarth, to lead UK’s Foundation Model Taskforce
• EU’s new AI regulation looks past ‘existential risks’ to focus on everyday life, says Italian PM
• International organizations:
• UN council to hold first meeting on potential threats of AI to global peace
• Hiroshima AI Process: The G7’s New Effort to Harmonize AI Rules

Responsible AI

• Predicting extreme weather risk 3 weeks in advance – at 1/10th of the compute time. (More from NVIDIA and the Berlin Summit)
• Aligning superintelligent AI with human intent is the focus of OpenAI’s new team
• DeepMind wrote about an early warning system for novel AI risks
• Mechanical Turk workers are using AI to automate being human

Bias:

• Training on only synthetic data breaks models, describes Anthropic co-founder, Jack Clark
• Stable Diffusion’s text-to-image model amplifies race and gender stereotypes (in podcast here)
• AI-detectors biased against non-native English writers

Energy overhead:

• Huawei unveils 3 new designs for green data centers
• New approach trains LLMs in half the time

Where AI is going

AI firsts:

• Human trials begin for AI-discovered and designed drug
• 1000x faster at designing a working computer
• VC and former head of Google China launches Lingyi Wanwu to build Chinese LLMs‍
• Defense start-ups, the Department of Defense, and AI service contracts‍
• New foundation models (ex. Anthropic’s Claude 2, Inflection-1, MosaicML’s MPT-30B) compete with ChatGPT 3-3.5 and LLaMa. (Visual: Rise of LLMs like ChatGPT)

Society

• AI researchers worry the US and China will leave everyone else behind
• Latest Ernie Bot outperforms ChatGPT 4 in Chinese language, says Baidu; Korean LLM released by EleutherAI’s Polyglot Project
• AI reshapes how we think about life itself, believes Geoffrey Hinton. (More on this theme in God, Human, Animal, Machine by Meghan O'Gieblyn)

AI in highly-regulated industries:

• 90% accuracy standard required by Europe’s largest rail system
• AI may someday work medical miracles. For now, it helps do paperwork
• Generative AI, enterprise software, and financial services

Sign up for weekly updates in your inbox here. What's happening in AI is shaped by our Charter themes and Responsible AI work.