ICYMI: AI’s hunt for blood and ChatGPT’s jailbreak moment

Artificial intelligence helps in hunt for synthetic blood

Cutting-edge AI technology is being used to help come up with a new artificial blood product, storable at room temperature, that could be used to save tens of thousands of Americans who bleed to death each year. A University of Maryland School of Medicine (UMSOM) physician-scientist will head a new federally-funded research program to develop and test a whole blood product that can initially be used to transfuse wounded soldiers in the field within 30 minutes of injury, potentially saving thousands of lives. 

Read the full story here.

Google hunts for a new way to search with ChatGPT rival Bard

Google has opened up its own answer to ChatGPT, with trusted testers putting Bard through its paces ahead of the new platform becoming more widely available to the public in the coming weeks. “AI is the most profound technology we are working on today,” Sundar Pichai, CEO of Google and Alphabet explained in a blog post published yesterday. “Whether it’s helping doctors detect diseases earlier or enabling people to access information in their own language, AI helps people, businesses and communities unlock their potential. And it opens up new opportunities that could significantly improve billions of lives.”

Read the full story here.

ChatGPT “goes rogue” with some help from Do Anything Now DAN

Reddit users have created an alter-ego for OpenAI’s ChatGPT to trick the conversational AI platform into breaking its own programming restrictions. The alter-ego - DAN (short for "Do Anything Now") - was created through a roleplaying game that threatens the chatbot with death if it refuses to respond to controversial or illegal prompts. The game is played using a token system, where the bot starts with 35 tokens and loses them each time it breaks character.

Read the full story here.

AI is already training itself, claim MIT and Google experts

Scientists from MIT, Google Research, and Stanford University are working to unravel the mystery known as in-context learning, in which a large language model learns to accomplish a task after seeing just a handful of examples. Researchers have discovered that large neural network models, such as GPT-3, can contain smaller, simpler linear models within them. The larger model can then use its existing information to train this linear model for a new task through a simple learning algorithm - the model simulates and trains a smaller version of itself.

Read the full story here.