Shari Garmise

I’ve spent the last week in Richmond, Virginia freezing in this Arctic front impacting the east coast. Climate variability is the new normal. Not surprisingly, there have been advances in weather forecasting, brought to us by Google Deepmind’s AI’s GenCast.

Research published in Nature indicates that GenCast outperformed the European Center for Medium-Range Weather Forecasts, which is generally considered to be the best in the world. In a head-to-head, GenCast’s forecasts were faster and more accurate 90 percent of the time. GenCast can predicted weather 15 days in advance within 8 minutes for both daily and extreme weather conditions. It can also generate escape routes. So as these technologies improve, we will be better able to prepare for climate emergencies.

And my other signal today refers to the evolution of dataset consumption to train our emerging AI tools and companions. One of my worries is how tech companies are using our data to help machines learn, yet we have no voice in the process.

Or any remuneration.

Or any information on the process, which is opaque.

Our lives are fed into the machine and spit back out at us in new forms, understandings, and perhaps new worlds we will need to navigate. We all know Meta and Alphabet use the data they collect from our usage their products. Those and ChatGPT use language models.

But AIs are hungry little suckers and need more data to do more and different things. One of those things, is to build datasets that enable robots to evolve.

Niantic, creator of Pokemon Go, is using Pokemon Go data to feed geospatial machine learning, because robots need to understand the physical world.

Its been a busy week so I will stop at two signals today. I also don’t know if these give me hope, or scare me as they prompt me to contemplate more deeply about how fast we are racing into a unknown feature.

Images generated by AI.