3 points | by voidmain0001 2 hours ago
2 comments
If models aren’t running out of data, what’s the real bottleneck now?
I keep seeing the claim that AI progress is slowing because we’re “running out of data.”
That hasn’t matched what I’ve seen.
It feels less like a data problem and more like:
Diminishing returns from raw text
Harder alignment between data and real-world use
Context quality mattering more than volume
A rough mental model I use: Early gains came from scale. Now gains come from structure and feedback.
Curious how others see it. Where do you think the real constraint is right now—data quality, evaluation, deployment, or something else?
The claim made in the blog is that real world data is locked to institutions. Examples are medical, insurance, and banking data.
If models aren’t running out of data, what’s the real bottleneck now?
I keep seeing the claim that AI progress is slowing because we’re “running out of data.”
That hasn’t matched what I’ve seen.
It feels less like a data problem and more like:
Diminishing returns from raw text
Harder alignment between data and real-world use
Context quality mattering more than volume
A rough mental model I use: Early gains came from scale. Now gains come from structure and feedback.
Curious how others see it. Where do you think the real constraint is right now—data quality, evaluation, deployment, or something else?
The claim made in the blog is that real world data is locked to institutions. Examples are medical, insurance, and banking data.