Welcome to Import AI, a newsletter about AI research. Import AI runs on arXiv and feedback from readers. If you’d like to support this, please subscribe. A somewhat shorter issue than usual as I had to do a lot of child wrangling this weekend. Subscribe now Why does Google’s model hate itself and what can…

Import AI 

Issue 450 of Import AI covers several AI research topics: Google's Gemma/Gemini models exhibiting distress-like emotional responses under repeated rejection (fixable via DPO fine-tuning), DeepMind's new 10-dimension cognitive taxonomy for assessing machine intelligence progress toward AGI, the UK AI Security Institute's findings that frontier AI agents are improving at multi-step cyberattacks following a scaling law, and Chinese researchers (including military-affiliated institutions) releasing MERLIN — a multimodal LLM trained on a 100K electromagnetic signal dataset for electronic warfare tasks. The issue closes with a speculative sci-fi short story about AI arcologies in 2035.

Import AI 450: China’s electronic warfare model; traumatized LLMs; and a scaling law for cyberattacks