REVIEW 14 cited by
Calibration and Correctness of Language Models for Code
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
Signed reviews
read the original abstract
Machine learning models are widely used, but can also often be wrong. Users would benefit from a reliable indication of whether a given output from a given model should be trusted, so a rational decision can be made whether to use the output or not. For example, outputs can be associated with a confidence measure; if this confidence measure is strongly associated with likelihood of correctness, then the model is said to be well-calibrated. A well-calibrated confidence measure can serve as a basis for rational, graduated decision-making on how much review and care is needed when using generated code. Calibration has so far been studied in mostly non-generative (e.g. classification) settings, especially in software engineering. However, generated code can quite often be wrong: Given generated code, developers must decide whether to use directly, use after varying intensity of careful review, or discard model-generated code. Thus, calibration is vital in generative settings. We make several contributions. We develop a framework for evaluating the calibration of code-generating models. We consider several tasks, correctness criteria, datasets, and approaches, and find that, by and large, generative code models we test are not well-calibrated out of the box. We then show how calibration can be improved using standard methods, such as Platt scaling. Since Platt scaling relies on the prior availability of correctness data, we evaluate the applicability and generalizability of Platt scaling in software engineering, discuss settings where it has good potential for practical use, and settings where it does not. Our contributions will lead to better-calibrated decision-making in the current use of code generated by language models, and offers a framework for future research to further improve calibration methods for generative models in software engineering.
Forward citations
Cited by 14 Pith papers
-
Uncertainty-aware Generative Recommendation
UGR injects uncertainty into generative recommendation preference optimization via confidence-weighted rewards, difficulty-aware reweighting, and explicit confidence tokens, improving ranking accuracy and training stability.
-
From Noise to Knowledge: Interactive Summaries for Developer Alerts
CLARITY uses interactive inductive-logic rules to group static-analysis warnings that share structural or code-level features, and reports modest user-study gains in sensemaking.
-
Are They All Good? Evaluating the Quality of CoTs in LLM-based Code Generation
An empirical study of 1,023 CoT-code pairs shows that 76.4% of LLM-generated CoTs are low quality and that CoT correctness does not guarantee code correctness.
-
SQLens: An End-to-End Framework for Error Detection and Correction in Text-to-SQL
SQLens uses weak supervision over database and LLM error signals to detect clause-level semantic errors in text-to-SQL and fixes them iteratively, improving execution accuracy on BIRD and Spider.
-
Correctness Assessment of Code Generated by Large Language Models Using Internal Representations
Internal hidden states of code LLMs carry a usable signal about whether the generated code is correct, and a lightweight probing classifier on those states predicts test-passing with F1 around 0.7 to 0.8 across benchmarks.
-
Unseen Horizons: Unveiling the Real Capability of LLM Code Generation Beyond the Familiar
A code-obfuscation benchmark shows LLM code generation pass rates fall sharply when descriptions, code, and dependencies are rewritten to remove training-data familiarity.
-
VeCoGen: Automating Generation of Formally Verified C Code with Large Language Models
VeCoGen automatically generates and formally verifies loop-free C functions from ACSL and natural language specifications, solving 13 of 15 Codeforces problems with its default model.
-
When Models Know When They Do Not Know: Calibration, Cascading, and Cleaning
A confidence-based, training-free routing and cleaning framework improves accuracy-efficiency trade-offs and identifies mislabeled data in image and language benchmarks.
-
Are LLM Agents the New RPA? A Comparative Study with RPA Across Enterprise Workflows
In head-to-head tests on three standard automation challenges, RPA bots beat LLM computer-use agents on speed and reliability, while the agents needed far less development time.
-
Maximizing Confidence Alone Improves Reasoning
Reinforcement learning with a reward equal to the negative entropy of the model's own token distribution improves benchmark accuracy without any labeled answers.
-
Understanding the Fundamental Design Decisions of Retrieval-Augmented Generation Systems
RAG deployment must be context-aware: QA benefits from 5 to 10 retrieved documents, code generation has no stable optimal document count, and prompting helps code tasks far more than QA tasks.
-
HASHIRU: Hierarchical Agent System for Hybrid Intelligent Resource Utilization
A hierarchical AI agent framework that dynamically hires and fires specialist models and creates tools reports strong benchmark numbers, though its gains may come from tool use rather than the architecture.
-
Can You Trust LLM Judgments? Reliability of LLM-as-a-Judge
LLM judges are often internally inconsistent across seed variations, with McDonald's omega reliability scores mostly below acceptable thresholds.
-
A Survey on Uncertainty Quantification of Large Language Models: Taxonomy, Open Research Challenges, and Future Directions
A review that organizes LLM uncertainty quantification into token-level, self-verbalized, semantic-similarity, and mechanistic interpretability categories.
Discussion (0). Continue with ORCID to comment.