• Chaos Theory
  • Posts
  • 🥟 Chao-Down #189 Researchers break AI watermarks, LinkedIn sees AI disrupting the 4-year college degree, National Security Agency launches an AI security center, Adobe teases new photo editing tool

🥟 Chao-Down #189 Researchers break AI watermarks, LinkedIn sees AI disrupting the 4-year college degree, National Security Agency launches an AI security center, Adobe teases new photo editing tool

Plus, Google DeepMind unites researchers to build the ImageNet for robot actions

As mentioned in yesterday’s newsletter, I got to give a recorded talk at the AI Engineer Summit titled “Cooking with the Semantic Kernel: Recipes for Building Chatbots, Agents, and more with LLMs.” If you want to check it out, watch here!

-Alex, your resident Chaos Coordinator.

What happened in AI? 📰

Google DeepMind unites researchers in bid to create an ImageNet of robot actions (TechCrunch)

Researchers Tested AI Watermarks—and Broke All of Them (WIRED)

LinkedIn sees AI disrupting the value of the four-year college degree (Axios)

AI was asked for images of Black African docs treating white kids. How'd it go? (NPR)

National Security Agency is starting an artificial intelligence security center (AP News)

Adobe teases a new ‘revolutionary’ AI photo editing tool (The Verge)

A screenshot taken of Adobe’s new Project Stardust tool.

Always be Learnin’ 📕 📖

LLMs are Interpretable (Tim Kellogg)

How AMD May Get Across the CUDA Moat (hpcwire.com)

Decomposing Language Models Into Understandable Components (Anthropic)

Projects to Keep an Eye On 🛠

Pythagora-io/gpt-pilot: Dev tool that writes scalable apps from scratch while the developer oversees the implementation (Github)

Announcing Replit AI for All - Code completion and code assistance are now enabled by default (Replit)

Google Assistant with Bard: New generative AI features (blog.google)

The Latest in AI Research 💡

LightSeq: Sequence Level Parallelism for Distributed Training of Long Context Transformers (arxiv.org)

LLM Finetuning Risks - Fine-tuning Aligned Language Models Compromises Safety,Even When Users Do Not Intend To! (llm-tuning-safety.github.io)

TiC: Exploring Vision Transformer in Convolution (arxiv.org)

The World Outside of AI 🌎

Ancient Zombie Viruses Alarm Scientists Studying Warming Arctic (Bloomberg)

China targets 50% growth in computing power in race against U.S. (Reuters)

Report: Spotify is being used by criminal gangs to launder money in Sweden (The FADER)

New Google research: Decoding consumer decision-making (thinkwithgoogle.com)

A hidden system of exploitation underpins US hospitals’ employment of foreign nurses (qz.com)

One Last Bite 😋