Last Week in AI

News from the week of 20 April

Coming in a day late, AI spent the week guzzling water, rewriting laws, passing bar exams in secret, and racking up bills listening to you try to be polite. Catch up below.

Model Benchmarks & Breakthroughs

• OpenAI seeks to make its upcoming ‘open’ AI model best-in-class

• DeepSeek opens roles for ‘product and design’ as start-up keeps mum on new AI model

---

Safety & Misinformation

• AI secretly helped write California bar exam, sparking uproar

• Anthropic is launching a new program to study AI ‘model welfare’

• Science sleuths flag hundreds of papers that use AI without disclosing it

---

Governance, Policy & Power Plays

• In AI race against US, China is racking up real-world wins

• AI bigwigs urge AGs to block OpenAI's profit pivot

---

AI in the Wild: Embodied and Deployed

• ChatGPT burns tens of millions of Softbank dollars listening to you thanking it

• AI Drives Re-Engineering Of Nearly Everything In Chips

• Microsoft sees hybrid offices emerging as humans, AI agents team up to boost efficiency

• Chinese carmakers put brakes on autonomous-driving hype at Shanghai show

• Researchers use machine learning to engineer 'bespoke enzymes' for gene editing

• AI Is Spreading Old Stereotypes to New Languages and Cultures

---

Fringes of AI

• ESA and IBM launch AI model with ‘intuitive’ understanding of Earth

• The potential of all-AI labs

• Ex-NSA chief warns AI devs: Don’t repeat infosec’s early-day screwups

• AI is pushing the limits of the physical world

---

Platforms, Hardware & Scaling

• Within six years, building the leading AI data center may cost $200B

• TSMC shows off 1.4nm chip tech that will appear in future iPhones and other devices

• Even the U.S. Government Says AI Requires Massive Amounts of Water

• Intel AI Trick Spots Hidden Flaws in Data Center Chips