GOOGLE’S QUIET AI GAMBLE

Many commentators are calling Google’s newly revealed AI path confusing. And underwhelming. But for those leaning in, it is anything but.

The muted reaction is, perhaps, understandable. The models launched at Google I/O were catch-up not leap-ahead. The apps weren't ready. And Gemini Pro was punted to June.

Then AI legend Andrej Karpathy joined Anthropic the same day - drawing 25m+ views on X - and pulled considerable focus.

It's also not usually great news when procurement implications make headlines. Now costing 3x from the previous, Gemini Flash is no longer a budget model. And early reviews include the words ‘verbose’ and ‘skittish’ as well as ‘expensive’. Triply poor in combination.

Meanwhile, the AI-and-jobs debate took another high-profile turn, disipating even more attention. James Manyika, who heads Google's societal-impact team, directly challenged the dismal forecasts from Anthropic and Microsoft. He offered to 'take the bet' against their dramatic white-collar job loss predictions.

And yet, what will the world remember in a year? Perhaps a far bigger bet being made by Google’s AI lead Demis Hassabis.

He posited that AGI is 'just a few years away'. And then framed Gemini Omni, a world model that simulates reality rather than predicting text, as the centrepiece of Google’s year.

Hassabis’ bet is that generally intelligent models need to understand physics and causation. And the early evidence of his bet's outcome, while nascent, can already be seen - today - within Gemini's new video generation and editing capabilities.

Will that bet pay off? Well, the big guy's all in.