21 posts
Federal law is published in Dutch and French. Regional codes exist in one language only. Rulings follow the language of the procedure. A monolingual AI tool misses half the corpus.
When Belgian tax sources disagree, the worst thing an AI tool can do is pick one and act confident. Here's what honest uncertainty looks like.
Accuracy claims without published metrics are marketing. Here's what it takes to measure legal AI honestly, and why the industry avoids it.
When Belgium's programme law amends 47 provisions overnight, what happens inside the AI tool you rely on? A transparent look at the ingestion pipeline.
Stanford found that premium legal AI tools hallucinate 17-33% of the time. But the most dangerous finding wasn't the hallucination rate — it was misgrounding.
First-stage retrieval finds 100 matches. Reranking identifies the 5 that actually answer your question. Here's why that distinction matters for legal AI.
Keyword search finds exact article numbers. Semantic search finds related concepts. Hybrid search does both — and the difference is measurable.
The AI Act entered into force on 1 August 2024. Some provisions are already active. Most articles confuse provider obligations with deployer obligations — here's the distinction that matters for your practice.
AI adoption in accounting firms surged from 9% to 41% in one year. McKinsey says 44% of legal tasks are technically automatable. The World Economic Forum lists bookkeeping among the fastest-declining roles. Here's what this actually means for Belgian tax professionals — and what it doesn't.
A search engine finds text. A knowledge graph navigates relationships: which article amends which, which ruling interprets what, which exception overrides the rule. Belgian tax law is a web of cross-references — and a knowledge graph is the map.
Before your legal AI tool can answer a question, it has to cut the law into pieces. The way it cuts determines whether the answer includes the exception that changes everything — or misses it entirely.
Harvey users report 37 hours per month. Thomson Reuters claims 63% reduction. We looked at the real numbers for Belgian tax research — and built a conservative estimate that doesn't require you to believe marketing claims.
Belgian tax law changes twice a year minimum. If your AI tool can't tell 2019 from 2026, its answer isn't wrong — it's for the wrong year. Here's what temporal versioning is and why it matters.
Chat interfaces feel modern but produce ephemeral, indefensible output. Professional tax research needs structured answers you can file, reproduce, and defend.
LLMs overestimate their own correctness by 20-60%. Confidence scoring doesn't fix that problem — it makes it visible. For tax professionals, that visibility is the difference between a research tool and a guessing machine.
Fine-tuning memorizes yesterday's law. RAG looks up today's. For Belgian tax professionals, this architecture choice determines whether your AI tool is current or confidently outdated.
Most firms start by asking 'how accurate is it?' That's question 10 on this list. Here are the nine questions you should ask first — and why they matter more for professional tax work.
The AI industry obsesses over accuracy benchmarks. For tax professionals, verifiability is the metric that actually protects you.
How retrieval-augmented generation works, why basic RAG still hallucinates, and what a search-RAG fusion architecture adds for tax professionals.
Why language models invent legal citations, what makes Belgian tax especially vulnerable, and three defenses that actually work.
Three tax questions, two AI tools, one clear lesson: for professional research, verifiability beats confidence.