Scaling laws are the finding that a model's performance improves predictably as its size, its training data and the compute spent on it increase. Emergence is the finding that particular abilities appear abruptly along that smooth curve. The two observations shape almost everything about how these systems are built and why they are hard to test, and both come from research published between 2020 and 2023.
The sections below describe each observation, then follow what the pair means for evaluating a model and for a product team choosing one.
Scaling laws
Model performance improves predictably as parameters, data and compute increase, and the relationship is smooth enough to extrapolate. The finding was published by OpenAI in 2020 and refined by DeepMind in 2022, which showed that most large models of the time had been trained on too little data for their size and that a smaller model trained on more data would do better for the same compute. A laboratory can choose a budget and forecast the loss it will reach before spending it.
That predictability is why the last several years looked the way they did. It turned capability into a planning exercise, and it justified spending at a scale that would otherwise be reckless.
The important qualification is what is being predicted. Scaling laws predict loss, the model's error at predicting the next token. They say nothing about which abilities that loss corresponds to.
Emergence
Individual abilities do not improve smoothly along with loss. Many are absent, absent, absent, and then present, across a relatively narrow band of scale. Multi step arithmetic and following an instruction format behave this way, and a 2022 paper catalogued more than a hundred such abilities.
So a laboratory can forecast the training curve accurately and still be unable to say what the finished model will do. That gap between a predictable aggregate and unpredictable specifics is the safety problem in this area.
There is a genuine argument that some of this is measurement. A 2023 paper showed that a task scored strictly right or wrong shows a sudden jump where a partial credit measure over the same outputs shows a steady climb. The model was improving all along and the metric only noticed at the end.
That does not remove the practical difficulty. Whether the underlying ability was gradual or not, the point at which it becomes usable, and therefore dangerous, still arrives suddenly from the outside.
What follows for evaluation
Three consequences follow for anyone testing a model.
The abilities to test for cannot be listed in advance. A model may have an ability nobody anticipated, so an evaluation covering the abilities somebody thought of is incomplete in a way that cannot be bounded.
Testing has to be repeated at each scale. A capability absent in the smaller model says nothing about the larger one, so results do not carry forward across a version.
Elicitation matters more than the model. The same weights can appear incapable and capable depending on prompting, tools and attempts, which is why a capability finding is really a finding about a model and a method together.
What follows for a product
Two practical things follow for a product team.
A model upgrade is a new system. Capabilities appear between versions, so the evaluation set has to be rerun, and that includes the adversarial cases.
Bigger is not automatically better. Scale buys breadth. On a narrow task with clear inputs, a small model often matches a large one at a fraction of the cost and latency, and the burden of proof belongs with the expensive option.