Welcome back to the AI For Humans newsletter!

Remember Bard? Google’s first big attempt at a public-facing AI chatbot was one of the biggest failures in modern AI history.

The company who had quite literally invented the transformer, THE piece of research tech that opened the door to ChatGPT & the LLM revolution, face-planted in their first attempt at commercializing it.

Go watch our criminally underseen LLM dating video from <checks notes> June of 2023 (?!) to see how bad it really was.

How much difference a few years makes:

So how did Google go from Bard’s flameout to this recent breakout?

Three big shifts stand out:

Get the Band Back Together: Reunite The Difference Makers

One of the under-reported parts of this comeback is cultural.

After Bard’s failure, Google realized it couldn’t keep running AI like two competing grad-school projects. Google Brain in Mountain View and Google DeepMind in London were brilliant but siloed. In 2023, Sundar Pichai finally merged them under Demis Hassabis, creating a new Google DeepMind. That meant one leadership chain, one roadmap, and one mission: make Gemini work or else.

Also, Google co-founder Sergey Brin himself returned to work. When one of the world’s richest and most famous tech founders shows up in the office, it sends a message.

But the “band reunion” went even further. In 2024, Google brought back Noam Shazeer, one of the literal co-inventors of the Transformer, who had left to start Character.AI.

Don’t underestimate the power of reunions. Sometimes progress isn’t about fresh faces, it’s about creating the conditions where your best people can do their best work again.

Don’t Be Afraid to Torch the Money Printer (Kinda)

It’s impossible to over-estimate the importance of Google’s Adsense to the story of Google as a company. The initial search algo is THE product of the last 25 years (outside of maybe the iPhone) but AdSense and its remarkable ability to make money is how Google became the Google we know today.

But Google’s search product got worse as the business people took control & started filling up results with more and more ad links. And, when ChatGPT appeared, it wasn’t hard to see a world where Google, the search engine, would become obsolete.

This issue is famously known as the innovator’s dilemma (Netflix vs Hollywood is another good example) and most often results in the established company getting beat by the scrappy start-up.

But Google, clearly after the Bard snafu, saw the future coming and pressed the gas down hard, turning the massive company towards AI across the board. It wasn’t just releasing a fully capable competitor in Gemini (now, some would say better product) it was implementing Gemini-powered AI search on their results page.

None of this really helps AdSense make more money right now… but if the shift happens to chatbots that seems inevitable, they’re now set up to compete.

Don’t cry for Google though. Like Meta, they’re making a ridiculous amount of money right now and, unlike their biggest competitor, they don’t need to keep raising cash.

Make It Delightful & Novel: The NanoBanana Effect

Maybe most importantly, Gemini’s brand new image model, NanoBanana has given Google something it hadn’t had for a long time: joy.

Google continues to make new and innovative progress on real world AI research but normal humans have a tendency to blow right past that.

NanoBanana finally gave Google its mainstream moment.

That matters. Shareable output is distribution. Every “how did you make this” comment pulled new users into Gemini.

Usefulness gets you trust; delight gets you growth. NanoBanana proved Google could do both in one place.

Google Might Just Own The Future Of AI

When you look around at the major players in the AI race, it’s hard to pick anyone other than Google to get to superintelligence first.

Meta is spending a fortune but we have no idea how that will turn out, OpenAI needs to make money & Anthropic seems to be doubling down on coding (for now). Don’t count out OpenAI co-founder Ilya Sutskever but it might be a few years until we hear from him.

Google, with its vast revenues & research background, is starting to look more and more like the company that’s gonna make i all the way. And considering how close they were to absolute & total failure with Bard, it’s a pretty fascinating case study.

That’s it for today. See you on Friday for the podcast!

- Gavin (and Kevin)

THIS WEEK’s AI For Humans: OpenAI’s Betting Big Bucks on AGI & Beyond 👇

We 💛 This: Seedream 4.0 = Another A+ AI Image Model

There’s a new competitor to Nano Banana that’s worth checking out: Bytedance’s Seedream 4.0. It’s crazy good & worth spending time playing with.

Unlike NanoBanana, Seedream 4.0 doesn’t exist in one of the main LLMs, so you have to find a different place to use it (like Higgsfield, Fal or one of the many other AI front-ends) but as you can see in some of the workflows below:

• Costume Design Iterations and one via GLIF here.

• Put Yourself on the Met Gala Red Carpet

• Generate Hand Written Recipe Cards

• Create Cinematic Action Shots (Great For AI Video starers)

• Generate Dynamic Angles For Even BETTER Cinematic Action Shots