Intel / post

Legal AI Analysis

Source: x.com
Author: Suhas Baliga
Date: Indexed Apr 15, 2026

Context

Suhas Baliga argues that a common legal-AI design can miss an exception: it splits documents into short passages, retrieves those whose wording most resembles the question, and may return a liability cap without the carve-out that qualifies it. Baliga's article proposes a document-by-document workflow instead: build a structured index of each document, use it to pick the relevant documents, read each in full for the question, then combine the findings with citations and contradictions preserved. The intended benefit is an answer whose reasoning stays traceable to the clauses that govern it. The article describes the design; it reports no measured accuracy comparison.

Key takeaways · AI-distilled

Vector search ranks by wording, not authority. Ask about termination rights and you get every paragraph containing the word “termination,” while the clause that actually controls the answer, titled something bland like “Consequences of Default,” never surfaces.
Top-k, the number of fragments retrieved, is a hardcoded guess. Set it low and the controlling clause gets cut; set it high and the model wades through irrelevant text. Nothing in the answer tells you which happened.
The synthesis step sees a scrambled collage. The model cannot reliably tell which fragment came from which document or with what authority, so it cannot tell you the master agreement and the work order contradict each other — it never held them as two objects.
The map-reduce alternative reads each document end to end and stores a structured map of it — parties, obligations, caps, deadlines — then answers in two passes: pick the documents that matter, then read those in full. Selection happens per document, not per fragment.

Terms in this piece · Glossary

RAG — Retrieval-augmented generation — fetching relevant documents first and pasting them into the model's context so it answers from your data instead of memory.
chunking — Splitting documents into passages small enough to embed and retrieve individually — the step that quietly determines whether retrieval works at all.

Key quotes

“The AI answers confidently, and it answers wrongly, and the confidence is the dangerous part.”

“On a legal matter, plausible is not the same as correct, and it is very far from defensible.”

“A contract is not a pile of paragraphs.”

“This is fine if you are building a chatbot. It is not fine if you are closing a transaction, advising on a dispute, or running a diligence exercise where the cost of a missed clause is measured in millions.”

Read the source x.com