Can LLMs Perform Synthesis?

2026-03-24T08:52:06Z540322082fd23d7d8ef39ca6b3b15afcf1f964a9fad211a51542dc35d680934d
AI hardwareErlangLLM agentsLLMsMINISASyGuSabstract machinesaccess controlchoreographic programmingdistributed protocolsdistributed transactionsfault toleranceinstruction set architecturepermissionsprogram synthesisprogram verificationreactive synthesisreconfigurable acceleratorsrepresentationruntime systemsset-theoretic typesstatic typessymbolic regressionsymbolic synthesistype checking

What happened

Collection of recent programming-languages and systems research from arXiv covering: (1) an empirical comparison of LLMs (Qwen-32B coupled with a verifier, GPT-5) against state-of-the-art symbolic program synthesis tools across domains (LTL/reactive synthesis, SyGuS, distributed protocol and recursive-function synthesis) showing symbolic tools solve more benchmarks and are faster; (2) Accompanist, a resilient runtime for choreographic programming that enables decentralised saga-style transactions via sidecars and provides formal correctness proofs under modest assumptions (determinism, idempot

Why it matters

A reviewed impact interpretation has not been published for this record.

Evidence and limitations

Source ID
arxiv_cs_pl
Record identifier
540322082fd23d7d8ef39ca6b3b15afcf1f964a9fad211a51542dc35d680934d
Enrichment time
2026-03-24T08:52:06Z
AI-assisted enrichment
Yes

This record may overlap with other records. Its enrichment can be incomplete or wrong, and machine assistance was used. Validate consequential decisions against the linked source and your own environment.