Neurosymbolic AI vs LLMs
What is neurosymbolic AI vs an LLM? A language model alone proposes an answer in its own text. A neurosymbolic system also runs a program or a formula and checks the result against facts. XERJ Query-1 is one: it compiles a prompt to a verifiable search program, then checks the answer against stored facts.
TL;DR A language model alone proposes an answer in its own text. Neurosymbolic AI also runs a program or a formula and checks the result against facts. XERJ Query-1 is one: it compiles a prompt to a verifiable search program.
A language model alone, or a checked program
The model proposes, and the answer stays inside it
A language model is strong at reading a question and drafting a reply. If the final number lives only in that reply, you cannot rerun a hidden step.
The model proposes, the engine checks
A neurosymbolic system still uses a neural part to propose a program or a formula. A symbolic part then runs it over facts or over words on a page, so you can read the program and run it again.
Query-1 and Reader-1
XERJ Query-1 compiles a prompt to a verifiable search program, then checks the answer against stored facts. Reader-1 writes a formula over a document, and the value is calculated from the words on the page.
Act as a coding agent. Read https://xerj.org/llms.txt, install XERJ, start a local node, index one folder I point you to, and run one search so I can see a cited passage.
How they differ
A language model is strong at reading a question or a document and proposing a next step. It is a poor place to hide the final number, because you cannot rerun a hidden step.
A neurosymbolic system still uses that neural strength. It also has a symbolic engine that runs a short program, or a formula, over stored facts or over words on a page. You can read the program, and you can run it again.
The comparison is not neural versus not neural. It is proposal only versus proposal, then a check.
How XERJ uses it
XERJ Query-1 is a neurosymbolic model that compiles a prompt to a verifiable search program, then checks the result against facts. It ships in rc.101. The product page is XERJ Query-1. The write-up is the Query-1 post.
XERJ Reader-1 writes a formula over a document, and the value is calculated from the words on the page. If the documents do not support an answer, it replies Not Found. See grounded answers.
On 12 November 2026, Model.Surgery joined XERJ to build neurosymbolic systems. The published line from that post is: models that propose, programs you can read, and facts that check the answer.
The closer look at running a program on symbols is symbolic execution for LLMs. The general concept page is What are neurosymbolic models?.
History and research
A recent survey is d'Avila Garcez and Lamb, Neurosymbolic AI: The 3rd Wave (arXiv 2020, later in Artificial Intelligence Review 56, 2023). That paper is a survey of the field. It is not a description of Query-1 or Reader-1. XERJ's product wording is the source for how the idea appears in this product.
Guides
Read how XERJ builds it: Model.Surgery joins XERJ, XERJ Query-1 and XERJ Reader-1.
Related terms
Neurosymbolic models, symbolic execution, and grounded answers.
FAQ
What is neurosymbolic AI?
It is a system that pairs a neural model, which proposes, with a symbolic engine, which runs a program or formula and checks it against facts or against words on a page.
How is this different from an LLM alone?
A language model can draft a program or a reply. In a neurosymbolic system the symbolic part runs that program and checks the result, so the final number is not hidden in the model.
How does XERJ Query-1 use this?
Query-1 compiles a prompt to a verifiable search program over stored facts, and ships in rc.101. Reader-1 writes a formula over a document. If the evidence is missing, the answer can be Not Found.
Where is the broader concept page?
The concept page for the models themselves is What are neurosymbolic models? This page is the comparison with a language model alone.
Related
- What are neurosymbolic models?
- Symbolic execution for LLMs
- What are grounded answers?
- XERJ Query-1
- XERJ Query-1
- Model.Surgery joins XERJ
- XERJ Reader-1
- BM25 vs vector search · same intent: Science
- Hybrid vs vector vs semantic search · same intent: Science
- What is agentic memory? · same intent: Science