AI progress is slowing because of an overabundance of useless data—there simply are too many.

Key takeaway: moving from ChatGPT to “real” robots requires not just large datasets but high‑quality, relevant information.

Fortune emphasizes that the AI industry stands on the brink of a new era—physical intelligence and models of the surrounding world. Such systems must be able to navigate real space: drive on roads, fold laundry, or assist in complex surgical procedures. For this they need not just “loadable” data but rich, multifaceted information sets.

Why data quality is critical
1. Redundancy and uselessness

In the AI hype, many startups (Scale AI, Surge AI, Mercor) aim to collect as much data as possible. This leads to a buildup of massive amounts of “empty” files that do not aid model training. If researchers cannot filter them out, the potential of physical AI will remain underutilized.

2. Complexity of multidimensional tasks

Understanding the complex world requires even larger volumes of data, but they are extremely hard to obtain. Therefore engineers resort to modeling: creating virtual reconstructions of real scenarios to generate training examples for robots and autonomous vehicles.

3. Risks of low‑quality datasets

Poor data can lead to unpredictable results. An example is the discontinuation of OpenAI’s video app Sora: the model struggled with physics, hindering realistic predictions.

How to solve the problem
- Cleaning tools

Machine‑learning specialists need technologies that automatically delete excess data, analyze, normalize, and correct datasets. This will allow valuable information to be extracted from “noise”.

- Focus on quality

Currently the limiting factor is a lack of high‑quality data. Companies that first understand this and begin investing in its collection and processing will become leaders in creating truly working AI systems.

Thus, moving from text models to robots capable of acting in the real world is impossible without a systematic approach to training data quality. Fortune stresses that this very factor will be decisive for the future of physical intelligence.