Welcome back to the AI For Humans newsletter!

The AI world continues to churn & progress this week as OpenAI preps a new model release (see below) and Chinese AI giant Kling rolled out a full new suite of tools.

Oh, and we got video of what looks like the most terrifying new humanoid robot to date, the original footage of which was so unbelievable they had to release the BTS footage.

But one story continues to dominate mainstream AI news:

When will the AI bubble pop and who will go down with it?

While it seems that every significant news organization now has multiple reporters tracking the overwhelming investments made in AI and your parents can now probably reference the viral circular investment infographic, it’s rare that actual people who understand the edges of the tech weigh in on the topic.

That’s why it’s important to watch the following interview with Anthropic’s Dario Amodei from the NY Times’ Dealbook Summit.

Dario Amodei is one of maybe three people in the world (Sam Altman & Google’s Demis Hassabis as the other two) who lead the edge of AI research. The team at Anthropic just shipped Opus-4.5, a very good new frontier LLM model, and continue to move the space forward.

A former VP of Research of OpenAI, Dario and a few other people broke away from OpenAI in 2001 to form Anthropic mostly because they didn’t agree with how OpenAI was pursuing AI and their perceived disregard for AI safety. Over the last four years, the company has quickly become a significant competitor to OAI and, due to their focus on coding and enterprise AI use cases, has seen their valuation raise remarkably.

Love this newsletter? Forward it to one curious friend. They can join in one click.

However, it feels like even Dario’s tone towards the AI conversation has changed over the last six months or so. In this clip below, you can see him directly throw shade at OpenAI’s recent ‘yolo-ing’ of spending on compute. And, if you read between the lines, you can feel Dario kind of hedging on AI progress, especially when it comes to revenue.

One of the touchstones of the last few years in the world of AI was Dario’s October 2024 essay “Machines of Loving Grace” now known better as source of the phrase “a country of geniuses in a datacenter”.

In the essay, Dario argues that the upside of radical AI improvement is actually massive positive scientific progress via thousands of genius-level minds. And that all this progress towards AI is really and truly towards making humanity better and stronger via the ability of artificial intelligence.

I made a video about the essay from around that time & it’s an easy to way to get up to speed on what he writes if you’d rather watching something than read.

Either way, it’s a (mostly) optimistic reading of where we could be in a few years.

Since that time, we have seen some big improvement in AI models (another major point of AI inflection was OpenAI’s o3-preview model last December where we clearly understood inference scaling would work) but have we seen the massive scientific progress that Amodei projected here?

Not exactly.

To be fair, in the essay Dario mentions that we’d be seeing what he calls “SuperIntelligent AI” only as early as 2026 but it does seem to be moving slightly slower than expected. One of the biggest speedbumps he mentioned in the essay is the limited amount of available compute, which is exactly what Sam Altman’s massive ‘overspend’ is trying to allivate.

Still, it’s worth taking an incomplete look at some of the biggest AI advances in science over the last year to see we might be making progress towards this goal.

And what better way to do that than asking both ChatGPT Pro 5.1 AND Google Gemini 3 Pro.

Seems like a lot doesn’t it?

Not exactly a ‘country of geniuses in a datacenter’ yet but defintely gives you positive things to chat with the in-laws about over the holidays.

This is the future we’d like AI to bring and, really, what the promise of the entire field is about.

See you next week!

-Gavin

In this week’s AI For Humans: OpenAI’s Code Red & Their New Models Incoming👇

I don’t fully understand what sort of magic legal deals (or magic legal underpinning) Google’s Nano Banana Pro model has to allow celebrity image generation but they seem to have very much not nerfed the model when it comes to famous people.

Maybe it has to do with the embedded AI-watermarking or maybe they’re covered somehow on their other legal deals or maybe they’ve just decided to go for it. Whatever the answer is, it means you can do a lot more with real-life celebrity photos like selfies on famous movie sets.

The prompt that I saw here on Reddit and have posted below is a lot of fun but also might open the door to an ‘AI or not?’ conversation over the Christmas holiday. Or, maybe more importantly, might lead to a good ‘What are DeepFakes?’ conversation with Grandma.