B.S. Detectors + Why AI didn't disrupt the election

Hey there,

I hope life, work and the world of AI are treating you well! We're back with 3 big things in AI.

#1 – AI didn't disrupt the election 🇺🇸

Back when OpenAI's Sora video model was released, there was a ton of chatter about whether fake videos would affect the election. In September, 57% of American adults worried that AI would be used to create fake content around the election.

Well, Election Day is next week and there have certainly been a lot of disruptive events this year – but in my estimation, exactly zero of them were related to AI.

So, why did "AI election fakes" go from existential risk to nothingburger?

Basically, the general public has quickly developed a very good AI bullshit detector.

Remember the fake Pope photo to that fooled people? Well, even those of us who are not super tech-savvy learn fast. Now, people just think differently about what they see online, and AI deepfakes are simply much less of a threat than they were 12 months ago.

As a result, it's hard to imagine a scenario where a real-seeming celebrity video comes out and people believe it's genuine for more than a few minutes. The flipside: Trump tried to say that images of Harris's crowd sizes were faked by AI and nobody bought it.

Turns out the human brain's ability to detect B.S. is outpacing even the best AI models so far.

#2 – Anthropic's 'Computer Use' mode brings AI agents much closer to the mainstream

​Check out this video from Dave Guarino – he prompts Anthropic's Claude model to scan a website for certain data, then it takes over his computer, controls his mouse and does all the work!

Right now, our students are building a lot of awesome workflows for clients. Getting good at 'computer use' AI seems like the next frontier for building even more complex automations more quickly. And it will definitely move "hiring a virtual assistant" closer to obsolescence.

Here are the official docs on the Computer Use beta test.

#3 - Who benefits the most from AI?

There's a new study out that restates the claim that AI is most helpful for lower-skilled workers. I appreciate the thesis here: since AI does a lot of work for you, it theoretically brings your "lower competence" employees up to a level of "moderate competence," which could show up as a large percentage change in their productivity or their quality of output at work.

That said, Hollywood editor Zack Arnold and I are somewhat skeptical of this claim. In practice, what we've seen is that people who are really good already (i.e. software developers or film editors at the top of their games) are 3x-ing or 10x-ing themselves with AI. This "removal of gruntwork" seems to actually push lower-competence workers out of the equation. For example, a software development firm needs far fewer junior developers today because of AI, but also the AI that currently exists is not quite sufficient to turn a junior developer into a senior developer.

🤓 By the way, have you subscribed to our YouTube channel yet? Lots more free tutorials coming soon. Please help me stop my 10-year-old son from making fun of me because we have fewer subs than the teenagers who livestream Minecraft all day!

That's all for today!

As always, if you have any questions or topic ideas for next week, just reply.

Have a great day,

– Rob Howard
Founder of Innovating with AI