What if the next model doesn't save you?
Reports of slowing gains from model scaling have unsettled the industry. For enterprise AI programmes, it might be the healthiest news of the year.
🧗 What if the next model doesn't save you?
The industry conversation this month has been dominated by one question: are the gains from simply training bigger models slowing down? Researchers are openly saying the easy wins are behind us, and attention is shifting to models that think longer at inference time rather than models that merely know more.
I have no inside view on whether the wall is real. But I've noticed something about how enterprises react to the question — and that reaction is worth examining.
In many AI programmes I've observed, there is a quiet habit of deferral: a use case underperforms, and the plan becomes "wait for the next model." Retrieval is weak — the next model will compensate. Data is messy — the next model will cope. Users don't trust the output — the next model will be so good they'll have to.
If scaling slows, that excuse expires. And honestly, I think that would be the healthiest thing to happen to enterprise AI in years.
Because the programmes that work have never depended on the next model. They depend on:
Retrieval that finds the right context, from data that has been made findable
Evaluation that catches regressions before users do
Workflows redesigned around the capability, not bolted beside it
Trust built through grounded, citable, reviewable outputs
None of that improves when a bigger model ships. All of it compounds when you invest in it.
The shift toward inference-time reasoning points the same direction: the frontier itself is moving from "what the model knows" to "how the system uses it." That has been the enterprise reality all along.
An audit worth running with your team this quarter: list your AI use cases that are underperforming, and mark the ones whose recovery plan is essentially "a better model is coming." Every mark on that list is a systems problem wearing a procurement costume. Fix it now — because the cavalry may be slower than the roadmap assumed.
Reading: https://techcrunch.com/2024/11/20/ai-scaling-laws-are-showing-diminishing-returns-forcing-ai-labs-to-change-course/
Last updated