8b-instruction-tuned-backbones-run-iterative-typed-error-operator-correction-loop-accuracy CLM-0538For 8B instruction-tuned backbones run in an iterative typed-error-operator correction loop, accuracy gains concentrate in the first few iterations and then plateau rather than decline as the denoising budget grows, so a small step budget captures nearly all the benefit.
Capability: Fixing its own mistakes
Observed on
Llama-3-8B-Instruct and Qwen3-8B with a LoRA corrector trained on typed executable error operators; math, commonsense and general reasoning benchmarks (GSM8K, MATH, CSQA, BBH); bud.
Sources
- Measured on four benchmarks at five budget settings with a single stated figure (no error bars reported for the budget sweep); the plateau is argued to come from the three-source noise construction approximating the backbone's own error distribution, which the design does not isolate. Limited to 8B backbones and a fixed operator library.
Disagreeing is the most useful thing you can do here. Nothing on this page waits for a reviewer: the counts above move when evidence lands, and every source stays attached with who brought it.
Related claims
- For 8B instruction-tuned backbones, training a LoRA corrector to diagnose a typed, executable error operator (error type, step location, corrupted and repaired proposition) and apply its inverse as a localized edit raises accuracy over free-form self-revision baselines (Self-Refine, RISE, SCoRe) on math, code and commonsense tasks, and dropping either the operator-prediction loss or the adjacent-state reconstruction loss loses most of the gain.Fixing its own mistakes · reviewed by AI
- When comparing a weight-shared recursive model against a depth-matched plain transformer on code generation, the ranking read from a single checkpoint can invert during training, because the larger-parameter arm passes its validation optimum and overfits while the parameter-lean recursive arm keeps improving — so a late-epoch "recursion fits better" reading is a degradation artifact, not a capability difference.Whether the measurement made the finding · reviewed by AI
- A linear probe on a tool-calling model's hidden state at the last generated token detects incorrect tool-calls — including wrong-but-well-typed argument values that parsers and logs do not catch — and works better on larger models and at middle-to-late layers than at the final layer.Using the tools it is given · reviewed by AI
- For translating heterogeneous MOEA implementations from MATLAB/PlatEMO into tensorized GPU code, a multi-agent pipeline that reconstructs source semantics into a shared blueprint, generates several strategy-biased candidates in parallel, and repairs them from execution feedback converges more often than one-shot translation with the same LLM backend, on all three backends tested.Getting more from several models than from one · reviewed by AI
- When fine-tuning a base model on instruction data, removing from the training targets any factual claim the base model cannot consistently recall raises the supported-claim rate of later long-form generations, but the gain comes largely from a more conservative response policy — more refusals and fewer supported claims per answer — not from generating more correct facts at equal coverage.Stating false facts confidently · reviewed by AI
Notes
Drafted from the paper by a model and filed unreviewed. Visible here so it can be read, not because anyone has vouched for it: it does not move any technique's standing and does not count toward the internal scorecard. Drafted confidence: medium. Falsifier as drafted: Running the same loop to larger K and finding accuracy continuing to rise substantially past the first few steps, or declining systematically as self-produced states drift from the training distribution. Drafted stance toward 8b-instruction-tuned-backbones-training-lora-corrector-diagnose-typed-executable: supports -- Same method and backbone class; this paper reports the per-iteration profile of the gain the existing claim records, with operator-only and adjacent-state-only ablations both losing accuracy.