AI with Kyle Daily Update 078

Today in AI: Stop Superintelligence statement + Clippy's back

The skinny on what's happening in AI - straight from the previous live session:

Highlights

📎 Clippy's Back: Microsoft Launches AI Avatar "Mico"

Microsoft unveiled "Mico" at their Copilot fall release - an expressive blob AI avatar that listens, reacts, and changes colours. And, more importantly, tap Mico repeatedly and it transforms into a modernised Clippy as an Easter egg!

Kyle's take: It's a cute little blob that's hard to hate - very Pixar! Smart move giving AI a face for the consumer market, but keeping it inoffensive.

Compare this to Grok's "Ani" who you can dress in lingerie for NSFW chats - Microsoft's going in a VERY different direction!

Interestingly, ChatGPT never gave their voices faces, probably smart - if they gave their Cove voice a face now, people would say "that's not MY Cove"

Probably bigger and more impactful news is that Microsoft's also adding Copilot to Edge browser instead of creating a separate one like Atlas. Sensible move - they have (some) market share, just add AI features. They're in a position of safety and can develop on their own timescale.

Expect Google to do the same with Chrome before too long.

🕵️ "Sex Spies" Target Silicon Valley Engineers

The Times reports Chinese and Russian operatives using "sex warfare" to seduce tech workers, even marrying and having children with targets. Security consultants claim sophisticated LinkedIn approaches from "attractive young Chinese women" have ramped up.

Kyle's take: The investigation is pretty anecdotal - basically one security consultant talking about getting LinkedIn messages from attractive women and then some follow up similar stories.

But I don’t doubt this is in play. Of course there's espionage - very likely America are doing the same, this is the game. But this article is just a series of anecdotes with a wild headline.

What's funnier is the Reddit and Twitter response - loads of single, sexually frustrated developers saying "oh no please don't target me!"

Iif this underlies anything, it's that AI is strategically important. Xi Jinping said years ago this is China's focus. Trump's AI Act earlier this year was explicitly anti-China. And Russia in the mix for some extra spice apparently (they’ve been quiet on the AI front).

Source: The Times investigation (no paywall)

✋ 30,000 Sign "Stop Super-intelligence" Petition

Heavy hitters including Geoffrey Hinton, Steve Wozniak, and oddly Prince Harry signed petition calling for prohibition on superintelligence development until "broad scientific consensus" on safety.

Kyle's take: 

I’m not one of the AI progress at all costs, move fast and break stuff accelerationists. I am, not too suprisingly, pro-AI. But I also think there are very serious issues that need to be discussed and worked out.

I just don’t think petitions and getting high profile celebs to sign a statement is particularly valuable.

Also…40,000 signatures? That's pathetically low for something supposedly 64% of people support. The signatories are a weird mix - tech legends like Geoffrey Hinton and Steve Wozniak, Glenn Beck, Steve Bannon, Kate Bush, Prince Harry and Meghan? Huh?

There is also the fact that slow down just isn’t pragmatic. China won't stop if America does. Companies won't stop if competitors continue. Individuals using AI will outstrip those who don't. These "nice to haves" about stepping back for discussion? They'll be crushed by the "need to haves" of continuing progress.

The only way to stop is full unilateral cessation. Something that humans aren’t very good at! The example that is trotted out here is the ozone layer problem which we did (as a species) manage to deal with. That one example is used a lot because it’s kinda the only one…

Equally the box is open - Hinton said open-sourcing models was like "allowing the public to buy nuclear weapons at Radio Shack." Too late now. We have powerful open source models out in the world which can be fine-tuned to get nasty.

Lots of people talk about Pandora’s box. But I think the better metaphor here is Prometheus stealing fire from the gods. Fire is technology, indeed the “first” technology in many ways. We can do wonderful and terrible things with fire - we can heat ourselves, cook food and bring light to darkness. We can also burn down the neighbouring village.

Member Question: "Can you replicate more complex products, not just simple tools?" (in reference to my recent duplication of SaaS tools to save on subscriptions)

Kyle's response: Depends on complexity! A prompt like "replicate Salesforce" isn’t going to work! But we can do smaller modules, parts of Salesforce. This is a better way to think about this opportunity.

The simple standalone SaaS tools - TypeForm, DocuSign, Feedly (RSS reader), document signing - these are all buildable in an afternoon. Most need minimal maintenance. These are ideally the sort of projects to focus on.

But if you really want to build something more complex thing about how it breaks up. Most complex systems are a network of smaller, simpler components. Build out these components step by step, make sure they work and then start to string them together. Just be aware that more complexity means a longer and more difficult job!

Want the full unfiltered discussion? Join me tomorrow for the daily AI news live stream where we dig into the stories and you can ask questions directly.

Streaming on YouTube (with screen share) and TikTok (follow and turn on notifications for Live Notification).

Audio Podcast on iTunes and Spotify.