Hybrid Symbolic + Neural Agents
Combining classical AI (planners, solvers) with LLMs for verifiable reasoning.
Beyond Pure LLMs
Pure LLM agents are flexible but unreliable on tasks requiring exact reasoning (math, planning, logic). Hybrid systems combine LLMs with classical symbolic tools to get both flexibility and exactness.
Examples of Symbolic Tools
- Calculator — exact arithmetic
- SAT solver — Boolean satisfiability
- SMT solver (Z3) — constraint systems
- Theorem prover (Lean, Coq) — formal proofs
- Classical planner (PDDL, STRIPS) — multi-step plans with guarantees