AIAny
AI Infra2026
Icon for item

AskChem: Claim-Centered Infrastructure for Chemistry Literature Synthesis

Indexes chemistry literature as provenance-bearing atomic claims and provides a faceted taxonomy, evidence graph, and REST/SDK/MCP APIs so researchers and AI agents can retrieve verifiable, claim-level findings across papers; live index contains 2.4M claims from 147K papers.

Visit Website

Introduction

Why this matters

Most literature-search tools return ranked documents, leaving humans and AI agents to find, verify, and synthesize specific results manually. AskChem flips the retrieval unit from the paper to the provenance-bearing claim: each paper is decomposed into small, typed assertions grounded by a source DOI and a verbatim quote or explicit evidence locator. That shift reduces citation hallucination for agent readers and makes it straightforward to assemble cross-paper answers that are directly verifiable.

Key Findings
  • Claim-level indexing at scale: the live index contains about 2.4M claims drawn from 147K papers (1925–2026), stored with full-text and vector indices to support fast retrieval and browsing. This makes fine-grained facts first-class search objects instead of buried sentences inside full papers.
  • Complementary retrieval structures: a stabilized faceted taxonomy for hierarchical browsing, an evidence graph that links supporting/contradicting claims, and source records with DOI and metadata; these structures let users traverse supporting evidence rather than chase whole documents.
  • Agent-first ergonomics and evaluation gains: AskChem exposes REST/SDK/MCP endpoints for agents; on a benchmark (AskChem-Bench) grounding a GPT-5.5 reader in AskChem raised resolvable DOI rate to 100% and substantially increased citation density versus unguided readers, demonstrating reduced citation fabrication and higher coverage of recent high-impact work.
  • Operational tradeoffs: the system favors verifiability and speed over exhaustive paper-level summarization; some complex experimental details remain easier to inspect in the original paper than in isolated claims.
Who It's For and Tradeoffs

Great fit if you need to assemble or verify quantitative findings across chemistry papers, instrument agent readers with provable evidence, or interactively browse a taxonomy of chemical claims. It simplifies workflows where traceable provenance and rapid cross-paper comparison matter (e.g., method/metric lookup, surveying catalyst performance, building grounded agent answers).

Look elsewhere if you need full-paper narrative synthesis, deep methods replication from unstructured methods sections, or maximal numeric detail tied to figures and tables without following evidence locators back to originals; AskChem is optimized for claim retrieval and provenance, not as a drop-in replacement for paper-reading during careful experimental reproduction.

Where It Fits

AskChem sits between document search engines and closed agentic research systems: more granular and verifiable than standard retrieval, more open and interactive than heavyweight closed pipelines. It complements RAG-style readers by supplying resolvable, source-anchored claims instead of whole-document snippets, making it particularly useful as an agent-facing retrieval layer.

Back

Information

  • Websitearxiv.org
  • AuthorsBing Yan, Gregory Wolfe, Stefano Martiniani, Kyunghyun Cho
  • Published date2026/07/30

Categories

Tags

More Items

Natural Language Processing Papers2026
Icon for item

Compile by Training: Turning Natural-Language Specifications into Local Neural Functions

Yuntian Deng, Pengyu Nie +1University of Waterloo, Harvard University

Converts a natural-language function specification into a reusable local neural function by using teacher models to synthesize examples and finetuning a small adapter for a compact interpreter. Achieves higher semantic accuracy (83.6% on FuzzyBench-Hard) at the cost of roughly one minute compile time; produces versionable PAW artifacts for local deployment.

NLPllmloraqwenpaper+5
Large Language Model Papers2026
Icon for item

Language Models Can Control Their Own Attention

Namgyu Ho, Huzama Ahmad +4KAIST AI, Google DeepMind

Introduces Declarative Attention (DA), a zero-shot protocol that has LMs declare which parts of long context to attend to during chain-of-thought, letting the runtime build dynamic attention masks and skip most KV-cache reads. Produces large token savings (up to ~52% on Gemma-4-31B) with modest accuracy loss.

llmlong-horizonvllmgemmaqwen+5
Natural Language Processing Papers2026
Icon for item

It Takes Two to Match: Co-Evolving Generative Retriever with Reinforcement Learning

Runpeng Dai, Kaili Huang +2University of North Carolina at Chapel Hill, Apple

Generates compact keyword sets for both queries and items with LLMs and matches them directly via an inverted index. Uses supervised fine-tuning to align keyword spaces, then alternates GRPO-based reinforcement learning on query- and item-side generators to co-evolve representations and maximize retrieval F1 while staying compatible with keyword-based infrastructure.

retrievalLLMRLsftbenchmark+2