Windows Copilot Newsletter #27 - Microsoft Copilot+; Totalled Recall; ScarJo vs SamA
Windows Copilot Newsletter #27
Microsoft recasts the AI PC as ‘Copilot+’; Microsoft demonstrates ‘Recall’ to privacy shivers; Scarlett Johansson makes ‘Her’ feelings known to OpenAI…
G’day and welcome to the twenty-seventh edition of the Windows Copilot Newsletter. Here we curate and present what we reckon are the the most significant stories in the rapidly-evolving area of AI chatbots. Another massive week of announcements, so let’s dive right in…
Top News
Microsoft launches Copilot+: At its developer-centric Build event, Microsoft unveiled a new PC platform, specifically designed to offer deep on-device AI capabilities. For the first time in forty years, these ‘Copilot+’ branded PCs no longer employ the x86 architecture - and will be coming from nearly every major vendor in June. Here’s what Microsoft has to say. And The Verge.
Will Microsoft need to recall ‘Recall’? During the Copilot+ announcement, CEO Satya Nadella demonstrated a new ‘Recall’ feature, which records a screenshot of your desktop every 5 seconds, creating a 3-month-long data trail that can then be searched in plain English using AI prompts. Useful - or a massive privacy intrusion? Here’s what Ars Technica says.
OpenAI shuts down 'superaligment’: Have a pesky AGI you need to contain before it goes the full Skynet? Good thing OpenAI has a ‘superalignment’ group thinking hard about mitigations! Or, um, did, until last week. The group is disbanded - and no one in that group can say why without losing all their stock options.
Ms Johansson has a few choice words for Mr. Altman: Ms Johansson was rightly irked that a new voice generated by ChatGPT sounded very much like her own voice in the Spike Jonez film Her. That led to a sudden withdrawal of that new voice by the AI firm. That’ll learn’em - never take on the woman who took on Disney - AND WON.
Top Tips
Pressing stop on Recall: Only one tip this week. It’s the one everyone has been asking for - here’s Microsoft telling us how to disable their new Recall feature.
Safely and Wisely
Schneier on prompt injection attacks: Godfather-of-computer-security Bruce Schneier writes an insightful post on prompt injection attacks - and how every AI chatbot is vulnerable to them.
I discovered ‘LLM Kryptonite’ - now what do I do? A few weeks ago the author of this newsletter discovered a simple prompt that reduced every chatbot (except Anthropic’s Claude 3) to gibbering idiocy. I tried to report this flaw to the model vendors - only to find that no reporting mechanism exists. Read the full story in The Register.
Longreads
Anthropic’s Responsible Scaling Policy: They’ve thought long and hard about how to mitigate ‘unintended consequences’ produced by the deployment AI models - and here’s what they’ve come up with as a strategy.
Khan Academy teams with Microsoft: Khan Academy’s ‘Khanmigo’ was one of the first ChatGPT apps that looked like a game-changer for students. In partnership with Microsoft, it’s being made available to every schoolteacher in the USA - for free. Lots in this post about how educators are working with AI.
‘De-Risking AI’ white paper - now out
AI offers organisations powerful new capabilities to automate workflows, amplify productivity, and redefine business practices. These same tools open the door to risks that few organisations have encountered before.
Wisely AI’s latest white paper, ‘De-Risking AI’, lays a foundation for understanding and mitigating those risks. It's part of our core mission to "help organisations use AI safely and wisely". Read it here.
More next week - though as we’re in a gap between Microsoft’s big event and Apple’s big event on 10 June, it may not be quite so frenetic.
If you learned something useful in this newsletter, please consider forwarding it along to a colleague.
See you next week!
Mark Pesce
mark@safelyandwisely.ai // www.safelyandwisely.ai