Windows Copilot Newsletter #15 - Gemini image fail; Copilot Power Automated; tipping ChatGPT
Windows Copilot Newsletter #15
Gemini's image generator fails; Copilot gets Power Automated; why you might want to offer ChatGPT a tip
G'day and welcome to the fifteenth edition of the Windows Copilot Newsletter, where we curate the most important news about the rapidly-growing field of AI chatbots. It's been another big week, so let's dive right in...
Top News
Google withdraws Gemini's image generator: After some images of 'racially diverse' senators (and, rather more disturbingly, Nazis) surfaced last week, Google pulled the plug on Gemini's image generation tool - already fully integrated into the new version of its AI chatbot - sending it back for some (lack of?) diversity training. Read about that here.
Microsoft makes an investment in Mistral: After purchasing a 49% stake last year in the for-profit arm of OpenAI (a corporation with a famously complicated ownership and control structure), this week the software titan hedged its bets with a minority stake in French AI startup Mistral. Mistral's latest 'Mistral Large' AI chatbot is already available through Azure's cloud services. Why is this happening? My guess is that the troubles last year at OpenAI led directly to this. Read about the investment here.
Copilot gets plugged into Power Automate: A tool favoured by Windows power users, Power Automate can be used to construct complex scripts that automate highly repetitive tasks - such as renaming files or reformatting data. The latest 'preview' builds of Windows 11 now bring Copilot's capabilities to Power Automate. This should mean you'll be able to tell Copilot to go off and build any Power Automate script - right now that it's "automation features for Excel, PDF manipulation and file management." Watch this space, and read about it here.
Microsoft gets principled: At this week's Mobile World Congress in Barcelona, Microsoft Vice-Chair and President Brad Smith published "Microsoft's AI Access Principles", laying out its responsibilities to be a good corporate citizen in the age of AI. While it's good to see Microsoft taking the initiative here, I'm reminded of the Russian proverb: "Trust - but verify."
Top Tips
Offer ChatGPT a tip? It's well-known that being polite and even enthusiastic can get better responses from AI chatbots. But can you bribe ChatGPT into giving your good answers? A researcher offered ChatGPT a range of rewards - and punishments - to determine whether either bribery or threats can be persuasive. Read their report here.
Useful Hacks from Anthropic: The makers of the Claude AI chatbot have shared a wonderful list of 'Useful Hacks' - and these should work with just about any other AI chatbot. Have a look at them here.
Safely and Wisely
Hallucinating Medical References: Research has found that when you ask an AI chatbot for references to back up its medical claims, it has a propensity to 'hallucinate' them. Just like a snake oil salesman. Read about that here.
Are Hallucinations Inevitable? That's the question posed in a paper published this week which purports to demonstrate that waves hands with lots of fancy theorem-proving yes, they are always going to happen. If true, we may be able to limit them - but will never be able to eliminate hallucinations completely.
Longreads
Khan's Accomplishment: This column in the Washington Post explores Khanmigo, a ChatGPT-based tutor that Khan Academy creator Sal Kahn has been working on for over a year. Is it the 'perfect tutor' for the hundreds of millions of students who don't have access to a personal tutor? Read and find out.
Anthropic Hits the Jackpot: In the post-ChatGPT world, there's a frenzy to be an investor on the 'ground floor' of the AI revolution. (Imagine if you invested in nVidia in 1998, as I advised my boss at the time!) The New York Times tells the story of Anthropic's latest funding round.
Understanding the Microsoft Copilot Pro Value Proposition
Last month, Drew Smith and I released our first Wisely AI white paper - designed to help organisations evaluate whether the features in Copilot Pro justify handing Microsoft a hefty AUD $45 per month per person subscription fee. Read or download the white paper here.
We're now hard at work on our next white paper - more about that next week, when we’ll be back with more AI chatbot news!
If you learned something useful in this newsletter, please consider forwarding it to someone else who might benefit.
Thanks!
Mark Pesce
www.markpesce.com // Need help with AI?