# Introduction
A mannequin that claims it’s 90% assured ought to be proper 90% of the time. When that relationship breaks down, you get a miscalibration downside. The mannequin’s scores cease telling you something helpful about reliability.
For giant language fashions (LLMs), miscalibration is widespread. A 2024 NAACL survey discovered that confidence scores diverge from precise correctness charges throughout factual QA, code era, and reasoning duties.
One other examine on biomedical fashions discovered imply calibration scores starting from solely 23.9% to 46.6% throughout all examined fashions. The hole is constant.
The usual answer in classical machine studying is post-hoc recalibration: match a easy perform on a held-out validation set to map uncooked confidence scores to better-calibrated chances.
Three strategies dominate: temperature scaling, Platt scaling, and isotonic regression. All three have been designed for discriminative classifiers, and making use of them to LLMs requires care.
# Measuring Calibration
The dominant metric is Anticipated Calibration Error (ECE). It teams predictions into confidence bins, computes the hole between imply confidence and the noticed accuracy in every bin, and averages throughout bins weighted by dimension. ECE = 0 is ideal calibration.
A reliability diagram plots confidence towards accuracy. A superbly calibrated mannequin sits on the diagonal. An overconfident mannequin sits under it: the curve reveals excessive confidence, however accuracy would not sustain.
A 2025 analysis of GPT-4o-mini as a textual content classifier discovered that 66.7% of its errors occurred at over 80% confidence — the canonical overconfidence sample.
ECE alone is more and more considered as inadequate. A analysis paper recommends pairing ECE with the Brier rating, overconfidence charges, and reliability diagrams collectively. A single quantity obscures significant variation in the place and the way a mannequin misbehaves.
# Why LLMs Complicate the Customary Setup
The three strategies we cowl assume a set output area. A classifier produces one chance per class, and calibration maps them to raised estimates.
LLMs do not work this fashion.
4 problems matter right here.
The output area is exponentially giant: sequence-level confidence cannot be enumerated. Semantically equal outputs could have very completely different token-level chances. Confidence disagrees throughout granularities; a analysis paper on atomic calibration confirmed that generative fashions exhibit their lowest common confidence in the course of era, not at the beginning or finish.
And plenty of LLMs solely expose top-k token chances by way of their API, so classical calibration approaches that depend on full logit entry want modification.
# Making use of Temperature Scaling
Temperature scaling divides the logit vector by a scalar T earlier than making use of softmax. When T > 1, the distribution flattens and confidence drops. When T < 1, the distribution sharpens and confidence rises.
T is match on a held-out validation set by minimizing adverse log-likelihood. The tactic provides one parameter, preserves prediction rankings, and is affordable to compute.
The authentic formulation focused DenseNet picture classifiers. For LLMs, temperature controls the chance distribution over the vocabulary at every decoding step, so the identical logic applies.
The issue is Reinforcement Studying from Human Suggestions (RLHF). Publish-RLHF fashions develop input-dependent overconfidence: the diploma of miscalibration varies throughout inputs, and a single T cannot account for that variation.
Common ECE scores above 0.377 have been documented for fashions like GPT-3 in verbalized confidence duties, and a 2025 survey confirms that RLHF-tuned fashions constantly overestimate confidence throughout the board.
Adaptive Temperature Scaling (ATS) addresses this instantly. ATS predicts a per-token temperature from token-level hidden options, match on a supervised fine-tuning dataset, as a substitute of utilizing a single fastened T. Researchers confirmed that ATS improved calibration by 10–50% with out hurting activity efficiency. For any RLHF-tuned mannequin, ATS is a stronger baseline than customary temperature scaling.
Customary temperature scaling nonetheless works properly for base fashions earlier than RLHF. When miscalibration is roughly uniform throughout inputs, a single T is usually sufficient to right systematic over- or underconfidence.
The issue is restricted to post-RLHF fashions, the place input-dependent overconfidence means a single T cannot right all inputs.
# Making use of Platt Scaling
Platt scaling matches a logistic perform over the uncalibrated scores: p = σ(A·s + B), the place A and B are discovered from a held-out validation set with binary correctness labels.
The sigmoid form provides a parametric mapping with two free parameters.
Platt scaling was initially developed for SVMs however generalizes to any system that produces a scalar confidence rating.
The 2-parameter match can also be data-efficient in comparison with isotonic regression: it might probably produce usable estimates from a smaller calibration set, which issues in deployment contexts the place labeled correctness information is proscribed.
In LLM contexts, Platt scaling operates over sequence-level or token-level confidence scores.
A paper on LLM-generated code confidence discovered that Platt scaling produced better-calibrated outputs than uncalibrated scores. One other examine on LLMs for text-to-SQL launched Multivariate Platt Scaling (MPS), extending single-variable Platt scaling to mix sub-clause frequency scores throughout a number of generated samples — constantly outperforming single-score baselines.
Two limitations are documented. First, international sequence-level Platt scaling is just too coarse for duties the place correctness is determined by native edit selections: a single sigmoid mapping cannot seize sample-dependent miscalibration patterns.
Moreover, Platt scaling can degrade correct scoring efficiency for sturdy fashions.
# Making use of Isotonic Regression
Isotonic regression takes the non-parametric route.
It learns a piecewise-constant, monotonically non-decreasing mapping from uncalibrated scores to calibrated chances utilizing the Pool Adjoining Violators Algorithm (PAVA). There is not any assumed form for the calibration perform, which makes it extra versatile than Platt scaling when the confidence-accuracy relationship is not sigmoid-shaped.
The piecewise-constant output adapts to any monotone form: linear, stepped, or concave. That adaptability is the primary motive isotonic regression tends to outperform Platt scaling in empirical comparisons.
The fee is overfitting threat on small calibration units. The mapping solely generalizes properly when there’s sufficient information to constrain it.
Empirically, isotonic regression outperforms Platt scaling.
A rigorous comparability throughout a number of datasets and architectures discovered that isotonic regression beat Platt scaling on ECE and Brier rating with statistical significance, utilizing paired t-tests with Bonferroni correction at α = 0.003.
In that examine, a Random Forest baseline improved from a reliability rating of 0.8268 uncalibrated, to 0.9551 with Platt scaling, to 0.9660 with isotonic regression. Each strategies might degrade correct scoring efficiency for sturdy fashions, however the isotonic edge held constantly.
For LLM multiclass settings, it has been proven that customary isotonic regression will be improved additional with normalization-aware extensions, constantly outperforming each OvR isotonic regression and customary parametric strategies on NLL and ECE.
The information requirement is the binding constraint. Isotonic regression’s benefit is actual, but it surely would not switch to low-data deployment eventualities.
# What the Literature Leaves Open
Three gaps are value flagging earlier than deploying any of those strategies.
The RLHF interplay has been studied just for temperature scaling. How Platt scaling and isotonic regression carry out on post-RLHF fashions hasn’t been systematically examined. ATS exists as a result of customary temperature scaling wanted an specific repair for this case. Whether or not the opposite two strategies want comparable extensions is an open query.
Most direct comparisons of all three strategies come from the overall machine studying calibration literature. LLM-specific benchmarks that check all three head-to-head are uncommon. The ICSE 2025 code calibration paper is without doubt one of the few, and its scope is proscribed to code era.
Calibration set dimension is an actual deployment constraint. Isotonic regression outcomes from papers assume datasets giant sufficient to constrain the mapping. In manufacturing with restricted labeled examples, the hole between isotonic regression and Platt scaling could shut or reverse.
# Conclusion
Temperature scaling is the appropriate place to begin for many groups. For base fashions with out RLHF, a single T typically does sufficient.
For RLHF-tuned fashions, change to ATS: the per-token temperature handles the input-dependent overconfidence {that a} international scalar misses.
Platt scaling is the sensible alternative when the calibration set is small or when calibration wants to fit into a bigger pipeline. It is data-efficient and easy to implement. The limitation is scope: it might probably’t seize miscalibration that varies throughout samples, and it tends to degrade efficiency for sturdy fashions.
Isotonic regression has the strongest empirical observe report of the three. Use it when the calibration set is giant sufficient to constrain the mapping with out overfitting, and pair it with normalization-aware extensions in multiclass settings.
The choice that comes earlier than all of those is what “confidence” means for the duty. Token chance, sequence chance, verbalized confidence, and consistency throughout samples may give completely different values for a similar output. A calibration technique utilized to the mistaken sign would not enhance reliability. Getting that definition proper is the prerequisite for any of the strategies above to work.
Nate Rosidi is a knowledge scientist and in product technique. He is additionally an adjunct professor educating analytics, and is the founding father of StrataScratch, a platform serving to information scientists put together for his or her interviews with actual interview questions from prime firms. Nate writes on the most recent traits within the profession market, provides interview recommendation, shares information science tasks, and covers every little thing SQL.
