Gemini's release blows up in Google's face

In its rush to assert AI supremacy, Google makes serious judgement error on Gemini demo

Happy Friday!

I’m coming to you from Los Angeles today, where I am speaking to members of the Academy tomorrow. As you can imagine, there's a palpable sense of collective anxiety about AI in the entire entertainment industry — but more on that later.

Google's Gemini: A Tale of Missteps

Today, I have a few notes on the most significant AI news of the past few days: the release of Google's Gemini and its bungled release via a heavily edited demo. (Top line: it made Gemini seem much faster, better, and more capable than it actually is.)

So, what looked like a slam dunk for Google quickly imploded within hours, once again revealing the intense political dynamics at play at the frontier of AI development.

But let's look at the model itself. As the first 'ground up' multimodal model, Gemini is seriously impressive. This is a significant achievement from the team at Google DeepMind, who have been 'balls to the wall,' aiming to surpass 'AI upstarts,' like OpenAI.

We knew multimodal was coming, and it's here just a year after ChatGPT’s release — and this should have been a significant moment for Google and AI development. Instead, it's all gone horribly awry thanks to 'demo-gate.'

At first glance, the misstep seems painfully obvious. Couldn’t someone (ANYONE!) in the comms team at Google have flagged that this would inevitably backfire? How could they have been so blind?

AI Pressure Cooker

The drive to excel is intense — especially at Google — so intense that it can clearly impair judgment.

After years of holding AI breakthroughs close to its chest - including with transformers/LLMs — Google learned a bitter lesson a year ago when OpenAI leapfrogged them and 'took credit' for LLMs with ChatGPT. It is desperate not to be the slow, boring incumbent anymore.

And to be fair, Google is narrowing the gap with OpenAI on frontier AI development. We are now a year out from the release of GPT-3.5, and the only contender to GPT-4 is Gemini. The competition is intense - existential even. Google could taste victory over its archrival — and that flavor seems to have blinded it to even obvious errors.

However you cut it, this is a bad look for Google. Given how heavily Google (and all others) are leaning into lofty AI ethics and safety proclamations, I am still asking myself how this demo even saw the light of day. Even worse that it was tweeted by Google CEO, Sundar Pichai, where it racked up millions of views.

Of course, this may all ultimately be a storm in a teacup. I have no doubt that the 'aspirational' (at best) or 'deliberately misleading' (at worst) nature of the Gemini demo will become reality sooner or later. But by jumping the gun, Google scored a massive own goal. Instead of the headline being about its significant contribution to AI, it has become the story as an 'untrustworthy' developer of AI.

‘Trust’ and ‘AI’ — I’m sure that will continue to be a troublesome little dynamic we will revisit sooner or later.

For now - namaste,

Nina