Let It Think, Then Lock It In

Kari Jaaskelainen

13 Jan 2026 — 1 min read

Let It Think, Then Lock It In

Large language models shine at free-flowing reasoning, but that flexibility makes outputs hard to trust and parse. Constrained decoding (e.g., forcing JSON) fixes structure, yet can choke off reasoning.

This paper proposes a simple middle path: allow the model to reason naturally until special trigger tokens appear, then switch to structured generation. You get the best of both worlds: rich thinking first, guaranteed machine-readable answers after.

Up to 27% accuracy gains vs. pure free-form.
Only ~10-20 extra tokens of overhead.
Works across classification and multi-step reasoning tasks.
Delivers consistent, parseable outputs for production apps.

Why it matters: If your app needs reliable JSON or another schema but you don't want to sacrifice reasoning quality, "think before constraining" is a practical drop-in strategy.

Paper: https://arxiv.org/abs/2601.07525v1

Paper: https://arxiv.org/abs/2601.07525v1

Register: https://www.AiFeta.com

#AI #LLM #NLP #MachineLearning #StructuredDecoding #Reasoning #JSON #ArXiv #Research

Meet GRACE: a moral governor for safer, more transparent AI

AI agents are getting powerful—so how do we make sure they do the right thing, not just the effective thing? Meet GRACE, a reason-based moral governor that keeps AI behavior aligned with human norms by separating moral reasoning from goal-driven decision-making. * Moral Module: uses deontic logic and explicit reasons

Why some fine-tuned LLMs miss phishing—and how to fix it

Not all fine-tuned LLMs spot phishing equally. A new study tests Llama 3.1 8B, Gemma 2 9B, and Mistral on high-stakes phishing detection—and uses SHAP and mechanistic interpretability to reveal why models do (or don’t) generalize. * Architecture × data diversity matters: Gemma 2 9B hits state-of-the-art performance (F1

AI that explains itself—by following the science

AI that explains itself—by following the science Black-box AI can be powerful, but it often can’t tell us why it made a decision. Concept Bottleneck Models (CBMs) try to fix that by predicting human-understandable concepts first, then the final answer. The catch: standard CBMs ignore domain-specific cause-and-effect and

Meet GenomAgent: A Team of AI Specialists for Smarter Genomics Q&A

TL;DR: Finding reliable answers in genomics is hard. GenomAgent turns one big AI into a coordinating team of specialists—and beats the current leader by 12% on a key benchmark. Genomic facts live across many databases. Standard chatbots struggle because they can’t flexibly query those sources. GeneGPT added

Let It Think, Then Lock It In

Read more

Meet GRACE: a moral governor for safer, more transparent AI

Why some fine-tuned LLMs miss phishing—and how to fix it

AI that explains itself—by following the science

Meet GenomAgent: A Team of AI Specialists for Smarter Genomics Q&A