Weekly Links: #AIStratLondon, Muse, and Election Deepfakes

This week has been packed with travel and organizing the AI Strategy & Practice Conference in London (huge shoutout to the team!), so this update is a little shorter, but I'll throw in a link to the conference wrap-up post, which is really worth a read:

"AI Strategy & Practice Conference: The work behind AI that works"

In the news this week: Anthropic works on LLM formal verification of software, a global alert for users of PHP, and we now have more war dogs.

On to the larger stories, just four this week:

• Meta’s Muse Is Going Gangbusters. But Where Is the Payoff? Sorry to say, but coverage like this will likely be a poor take. Yes, Muse is expensive to deliver; yes, it has no advertising business model (yet), but it's hard to put a price on being the first to scale into this user-behavior shift. In the late 90s, Google launched Gmail with 1GB of free storage, and everyone lost their minds. It looked like a foolish thing to do while everyone else was happily charging for 20 MB of storage. 25 years later, it's clear this was the launchpad for all of Google's productivity apps.
• Opposition deepfake video of Spanish PM’s wife sparks outrage before election. AI election videos made by some of the major parties in the upcoming Spanish national election are insane and border on illegal. AI deepfakes are now part of election cycles, but when the leading parties behave like this, why would anyone restrain themselves? I guess "human in the loop" isn't happening in the party leaderships. Coverage (in Spanish) of the video that kicked it all off.
• Inside the Black Box of AI. This is a nice, non-technical interview with one of Anthropic's mechanistic interpretability team (yes, that already sounds technical!). In essence, it dives deep into how LLM AI models work without any technical fluff.
• Google Cloud outage in UK eases after impacting digital banks, public sector websites. Google Cloud outages hit many high-profile services in the UK this week, including banks and other financial services. The effects were severe and led to calls for more failover and diversity in supply. Honestly, though, it is hard to imagine more than 3-4 major hyperscale clouds being viable. Certainly, large institutions have failover for services to other regions and clouds; however, they are often hard to activate depending on the exact nature of the failure. I'd honestly say that the fact that most major services stay up almost all the time is a big positive. The failover will be even more important with escalating AI cyberattacks, however.

Wishing you a great weekend.