Hallucinate Less by Thinking More: Aspect-Based Causal Abstention for Large Language Models

Kari Jaaskelainen

24 Nov 2025 — 1 min read

Hallucinate less by letting AI “think” about context

Large language models can sound confident while being wrong. This paper proposes Aspect-Based Causal Abstention (ABCA): a way for models to pause early and say “I don’t know” when their own knowledge looks unreliable.

How it works: before answering, the model probes different “aspects” of its learned knowledge—like discipline (medicine vs. history), legal context, or time period—and uses causal reasoning to test which aspects truly support an answer.

Type‑1 abstention: aspects disagree (knowledge conflict) → the model holds back.
Type‑2 abstention: aspects consistently suggest “not enough info” (knowledge insufficiency) → the model abstains.

By deciding to abstain earlier—rather than after generating text—ABCA reduces hallucinations, improves reliability on standard benchmarks, and makes decisions easier to interpret.

Why it matters: safer, more trustworthy AI for high‑stakes uses (health, law, news). Example: for “Who won the 2024 election?” conflicting time aspects would trigger abstention instead of a guess.

Paper: https://arxiv.org/abs/2511.17170v1

Paper: https://arxiv.org/abs/2511.17170v1

Register: https://www.AiFeta.com

#AI #LLM #Hallucination #CausalInference #ResponsibleAI #AISafety #NLP

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

Hallucinate less by letting AI “think” about context

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