Writing

Small-model behavior does not predict large-model behavior

Some capabilities appear discontinuously as model scale crosses a threshold. Below the onset point, prompt optimization is wasted work: you are in the wrong regime.

Teams build a prototype with a small model because it's cheap, then watch it fail in messy ways: ignoring tools, breaking output format, losing the thread on multi-step constraints. Weeks go into prompt hacks and brittle routing logic to paper over it. Then someone swaps in a larger model, and half of those hacks turn out to have been unnecessary the whole time.

What happened is not mystical. Some capabilities appear discontinuously as model scale crosses a threshold. Below that threshold, small-model behavior is not predictive of large-model behavior. You are in the wrong regime, and optimizing there is wasted work.

Why loss can be smooth while capability is not

Training loss averages over a massive corpus and tends to improve smoothly with scale. But production capabilities are measured as task-level outcomes: did it call the right tool, did it follow the schema, did it complete all steps without a catastrophic error. Those metrics are nonlinear functions of token-level quality.

Slight improvements in average token prediction can push a model across a threshold where multi-step trajectories stop collapsing. Smooth microscopic improvements produce sharp macroscopic behavior changes. That is the pattern of a phase transition.

Wei et al. (2022) documented a concrete instance: on 3-digit addition and subtraction, GPT-3 stayed at near-zero accuracy for several orders of magnitude of training compute, then jumped sharply above random performance at roughly 13B parameters (~2×10²² training FLOPs). Nothing about the smaller models predicted that jump; the task simply looked unsolved until it was solved.

Drag the model size, watch which tasks are in the wrong regime

Same model, three tasks of increasing difficulty. The onset point — where the curve actually bends — moves right as the task gets harder.

0%50%100%7B13B34B70B175B405B
Easy (tool call, 1 step): 91% — above threshold
Medium (3-digit arithmetic): 76% — above threshold
Hard (5-step tool sequencing): 46% — below threshold
At small sizes, all three tasks read "below threshold" — no amount of prompt tuning fixes that, because the capability isn't there yet. Drag right and watch easy cross first, then medium, then hard. Curves are illustrative; the onset pattern follows Wei et al. (2022).

Architecture can shift the threshold

You do not always need a bigger model. You sometimes need a simpler problem. RAG removes the need for parametric recall. Tools remove the need for in-context arithmetic. Constrained decoding removes format failure. Reranking removes multi-hop reasoning burden. Each of these moves the task into a regime where a smaller model becomes viable, by changing the problem, not by changing the model.

Emergence checklistIf you see these patterns, you are probably below the capability onset threshold.
SignalWhat it means
Outputs are either correct or unusable, no middle groundBimodal distribution typical of near-threshold behavior
High prompt sensitivity at small scale that vanishes at largerThe model is not failing the task; it is failing the regime
Rapid jump in quality over a small size increaseYou crossed the onset; behavior is now above the phase transition
One capability unlocking makes adjacent ones easierCompositional capabilities have a shared threshold

Evaluate stability, not a single run

Below threshold, lucky single-run passes are common. The model memorized a surface pattern or got a favorable sample. Evaluating stability means running multiple paraphrases, multiple seeds, and measuring pass@k rather than pass@1. What you are measuring is whether the capability is in the right regime, not whether it was lucky once.

That brittleness shows up the moment you push past a single test case:

Ethan Mollick on small models acing one problem but breaking on a related second one.

References: Wei et al., Emergent Abilities of Large Language Models (2022), Scaling Laws (Kaplan et al., 2020)