# AI

Изследвания и приложения на изкуствения интелект във финансите и счетоводството

89 публикации [Вижте всички етикети](https://beancount.io/bg/bean-labs/research-logs/tags) [Обратно към всички публикации](https://beancount.io/bg/bean-labs/research-logs)

12 юли 2026 г. 5 минути четене

## [FinRAGBench-V: Multimodal RAG with Visual Citations in the Financial Domain](https://beancount.io/bg/bean-labs/research-logs/2026/07/12/finragbench-v-multimodal-rag-visual-citation-financial-domain)

FinRAGBench-V finds top models reach only 20–61% block-level citation recall on financial pages. Multimodal retrieval beats text-only by nearly 50 points.

Mike Thrift

ai llm machine-learning

11 юли 2026 г. 6 минути четене

## [Can LLM Agents Be CFOs? EnterpriseArena's 132-Month Simulation Reveals a Wide Gap](https://beancount.io/bg/bean-labs/research-logs/2026/07/11/can-llm-agents-be-cfos-enterprisearena-resource-allocation-benchmark)

Only Qwen3.5-9B survives 80% of EnterpriseArena's 132-month CFO runs, while GPT-5.4 and DeepSeek-V3.1 hit 0%. Skipped ledger reconciliation causes the failures.

Mike Thrift

ai llm automation

10 юли 2026 г. 6 минути четене

## [WildToolBench: Why No LLM Exceeds 15% Session Accuracy in Real-World Tool Use](https://beancount.io/bg/bean-labs/research-logs/2026/07/10/wildtoolbench-benchmarking-llm-tool-use-in-the-wild)

WildToolBench finds no LLM exceeds 15% session accuracy on 1,024 real-user tasks, with hidden intent and instruction transitions the sharpest failure modes.

Mike Thrift

ai llm automation

9 юли 2026 г. 6 минути четене

## [LLM Confidence and Calibration: A Survey of What the Research Actually Shows](https://beancount.io/bg/bean-labs/research-logs/2026/07/09/confidence-estimation-calibration-llms-survey)

Verbalized GPT-4 confidence hits only \~62.7% AUROC, barely above chance. Uncertainty-aware finance agents need better calibration than that.

Mike Thrift

llm ai machine-learning

8 юли 2026 г. 6 минути четене

## [JSONSchemaBench: Real-World Schema Complexity Breaks LLM Structured Output Guarantees](https://beancount.io/bg/bean-labs/research-logs/2026/07/08/jsonschemabench-structured-outputs-language-models)

JSONSchemaBench finds coverage collapses from 86% on simple schemas to 3% on complex ones, so LLM structured output can silently emit non-compliant JSON.

Mike Thrift

llm ai machine-learning

7 юли 2026 г. 5 минути четене

## [FinMCP-Bench: Benchmarking LLM Agents for Real-World Financial Tool Use under MCP](https://beancount.io/bg/bean-labs/research-logs/2026/07/07/finmcp-bench-llm-agents-financial-tool-use-model-context-protocol)

FinMCP-Bench scores the best of six LLMs at just 3.08% exact match on 613 real MCP financial tasks, a 20× collapse from single-tool to multi-turn use.

Mike Thrift

ai llm automation

6 юли 2026 г. 5 минути четене

## [FinTrace: Trajectory-Level Evaluation of LLM Tool Calling for Financial Tasks](https://beancount.io/bg/bean-labs/research-logs/2026/07/06/fintrace-trajectory-level-evaluation-llm-tool-calling-financial-tasks)

FinTrace shows frontier LLMs pick the right financial tools (F1 \~0.9) but score just 3.23/5 on using the results, the step that breaks write-back agents.

Mike Thrift

llm ai finance

5 юли 2026 г. 6 минути четене

## [FinToolBench: Оценка на LLM агенти при реален финансов инструментариум](https://beancount.io/bg/bean-labs/research-logs/2026/07/05/fintoolbench-evaluating-llm-agents-real-world-financial-tool-use)

FinToolBench съчетава 760 реални финансови API инструмента с 295 изпълними заявки, за да оцени LLM агенти върху реални финансови задачи — разкривайки, че консервативният процент на извикване от 22,7% на GPT-4o води до по-високо качество на отговорите (CSS 0,670) от агресивния TIR от 87,1% на Qwen3-8B, докато несъответствието на намеренията надхвърля 50% при всеки тестван модел.

Mike Thrift

ai llm automation

4 юли 2026 г. 5 минути четене

## [OmniEval: Omnidirectional RAG Evaluation Benchmark for the Financial Domain](https://beancount.io/bg/bean-labs/research-logs/2026/07/04/omnieval-omnidirectional-automatic-rag-evaluation-financial-domain)

OmniEval scores the best RAG systems at 36% numerical accuracy across 5 financial task types, so ledger agents need validation before writing entries.

Mike Thrift

ai machine-learning llm

3 юли 2026 г. 5 минути четене

## [LLM Anomaly Detection Survey (NAACL 2025): Strong Taxonomy, Absent Tabular Coverage](https://beancount.io/bg/bean-labs/research-logs/2026/07/03/llm-anomaly-ood-detection-survey)

The NAACL 2025 taxonomy holds, but tabular coverage is absent. Finance AI teams must adapt vision-model methods themselves.

Mike Thrift

ai llm machine-learning

2 юли 2026 г. 6 минути четене

## [Found in the Middle: Calibrating Positional Attention Bias Improves Long-Context RAG](https://beancount.io/bg/bean-labs/research-logs/2026/07/02/found-in-the-middle-calibrating-positional-attention-bias)

Subtracting positional bias from LLM attention weights recovers up to 15 points of RAG accuracy when evidence sits mid-context, aiding finance agent pipelines.

Mike Thrift

ai llm machine-learning

1 юли 2026 г. 6 минути четене

## [Uncertainty-Aware Deferral for LLM Agents: When to Escalate from Small to Large Models](https://beancount.io/bg/bean-labs/research-logs/2026/07/01/redact-uncertainty-aware-deferral-llm-agents)

ReDAct defers from a small model to a large one only when perplexity signals uncertainty, cutting cost 64% at matching accuracy for agent workflows.

Mike Thrift

ai llm automation

[Започни безплатно→](https://beancount.io/sign-up/?next=%2Fledger%2F) [Документация](https://beancount.io/bg/docs)

## Where to look next

- [Sitemap](https://beancount.io/sitemap.xml) - all pages with hreflang alternates
- [LLMs index](https://beancount.io/llms.txt) - curated links for AI agents (also /llms-full.txt for full docs)
- [Documentation](https://beancount.io/docs) - every doc is also available as text/markdown via .md suffix or Accept negotiation
- [REST API reference](https://beancount.io/docs/api-reference) - interactive endpoints, schemas, authentication, and code samples
- [OpenAPI spec](https://beancount.io/openapi.json) - OpenAPI 3.1 spec for this website's public content endpoints; the product contract for ledger operations is at https://beancount.io/api-gateway/v1/openapi.json
- [GraphQL](https://beancount.io/graphql) - public, read-only content API over the blog and research logs (no credentials, no ledger data): 12 query fields over 23 types
- [MCP server](https://beancount.io/mcp) - Model Context Protocol at https://beancount.io/api-gateway/mcp
