What is it that you can "eat" all day, never get fat, and even get smarter?

Carbonoi
Dec 21, 2024
2 min read

Generative AI is currently feasting at a data buffet, where we collectively produce hundreds of millions of terabytes of data daily.

These models learn from vast amounts of information that developers scrape from various online sources (web scraping or data crawling). Most of the time, permission to use this data isn't formally requested. Now, data owners are becoming protective. Not only are lawsuits starting, but they're also beginning to prevent programs from scraping data, as seen with The New York Times suing OpenAI and Microsoft for copyright infringement of data from its website.

AI developers, and amateur users like us, have been on an adventure experimenting with this cutting-edge technology... but this party is counting down.

2024 genuinely sparked an AI craze during a time when there was no real oversight. But in 2025, there will be increasing pressure from investors to generate returns, and society will demand more transparency and accountability.

As the era of "free data" comes to an end, it will become clear who has built their systems on solid foundations and who has tried to cut corners.

From now on, developers must rethink their approach. They need to stop throwing massive amounts of data at models to learn from, data that is sometimes reliable, sometimes not, sometimes relevant, sometimes not.

Instead, they need to shift to teaching machines with less data (Data Minimization), but with high-quality data that is relevant to the model's intended use, prioritizing 'quality over quantity.'

Currently, Google has a team of experts working almost daily to analyze Bard's results using human-integrated feedback loops, incorporating human input to repeatedly train the AI model. This improves results without always needing to add new data.

Anyone who loves buffets knows that eating too much often leads to regret later. It's time for the world of data to adapt as well, because every piece of data obtained comes with an inherent cost.

References

https://sustainabilitymag.com/articles/msci-6-sustainability-climate-trends-to-watch-in-2025 https://www.spiceworks.com/tech/artificial-intelligence/guest-article/generative-ai-is-data-hungry/