This week we look at an important free offering from Microsoft, delve deep into the sands of time as AI decodes ancient tablets, and explore a new AI Alliance between Meta and IBM. Oh, and then there is Cicero. Let’s get started!

News

AI decodes ancient tablets of stone: A team of scientists from several German universities have trained an AI to decode ancient texts of over 5,000 years old. You can read about the discovery here.

IBM and Meta form AI Alliance with 50 partners: The two technology giants have collaborated to form an Alliance to advance safe, responsible development of open source AI. Read about it here.

Elon Musk says Artificial General Intelligence (AGI) within 3 years: in a far reaching, not to mention controversial interview, the tech magnate said that AI would allow us to write a novel as good as JK Rowling amongst other things, within three years. Read about the interview here.

Meta announce CICERO a new AI: a new AI has shown remarkable talent for playing the strategy game Diplomacy. The game requires the ability to negotiate, gain trust across many individual players. Find out about it here

AI reconstructs images from brain acivity: With around 75% accuracy, Japanese AI researchers have discovered a process of reconstructing images seen by volunteers using just their brain activity. You can read a tweet about it here and read the full techie paper here

Tool

Not a tool per se, more a free course from Microsoft teaching the basics of AI for beginners. If you want to learn more in a structured 24 module 12 week course, then this is a very good starting point. You can check it out here.

Prompt

Prompts are funny things. I read this week that if you tell an AI that you will give it a tip if it answers really well that you will get a better, longer response than the original non tip prompt! Whilst that is interesting, I think that prompting works best to a formula. I created a perfect prompt infographic which I’ve shared previously that was very formulaic and consistently works well. It follows the following form: Be specific > context > reasoning > goals & constraints > iterate. This is a pretty reliable formula.

Somebody I follow on Twitter/X came up with another principle which is Instructions, not questions. Here’s his post explaining all:

That’s me wrapped up for another week. See you on Saturday…