We are LLMQuant, an open-source community focusing on AI, LLM (large language model) and Quantitative Finance. We aim to leverage AI to investment research with feasible collection of techniques and solutions.
Live Last read · last published · next check
Generative verifiers turn reward modeling into next-token prediction, lifting GSM8K Best-of-N accuracy from 73.0% to 93.4% and revealing a powerful new path for scalable AI reasoning
AQuA reports a 2.50 out-of-sample Sharpe, but its real breakthrough is an architecture designed to stop autonomous research agents from manufacturing false alpha.
Man Group has pushed 15 to 20 models through a machine driven research pipeline and into a human investment committee.
PAT can compress hours of investment research into minutes.
A structural teardown of LLM driven factor discovery, the two architectural ceilings nobody warns you about, and the exact numeric thresholds at which you should switch the machine off
Skill routing, registries, permission layers, and MCP explained end to end, plus the arithmetic that quietly decides which agents survive contact with production
SESA couples zero-data self-play with an evolving skill memory, and the ablations reveal something uncomfortable about where agent capability actually lives
A new 63-page survey maps 117 methods across memory, tool use, and planning, and exposes the one number almost nobody bothers to report.
Eleven thousand stars in under a month, twenty five connectors, and an approval gate on every dangerous action.
Set up LLMQuant Data MCP once, and your AI agent can pull SEC filings, research papers, market prices, macro indicators, and prediction market data on its own. No glue code, no custom integrations.
