MemGPT: 92.5% multi-session recall vs 32.1% baseline
MemGPT's OS-style memory tiers push GPT-4 multi-session chat accuracy to 92.5% versus 32.1% fixed-context—needed for multi-year ledger agents.
#llm
LLM
Large language model research with applications in financial tasks
SWE-agent: How Interface Design Unlocks Automated Software Engineering
SWE-agent (NeurIPS 2024) introduces Agent-Computer Interfaces (ACIs) — purpose-built layers between LLMs and software environments — showing a 10.7-percentage-point improvement over raw shell access and 12.47% resolution on SWE-bench with GPT-4 Turbo. Interface design, not model capability, is the primary bottleneck for autonomous coding agents.
SWE-bench: Can Language Models Resolve Real-World GitHub Issues?
SWE-bench evaluates language models on 2,294 real GitHub issues across 12 Python repositories using execution-based tests; at publication, Claude 2 resolved only 1.96% of issues with realistic retrieval, establishing the de facto benchmark for coding agents and revealing retrieval and patch-length failure modes directly relevant to Beancount write-back agents.
CodeAct: Why Executable Python Code Makes LLM Agents 20% More Accurate
CodeAct (ICML 2024) replaces JSON tool-calling with executable Python code, improving GPT-4 agent success rates by ~20 percentage points on multi-tool tasks and reducing interaction turns by 30% — with direct implications for building reliable Beancount reconciliation agents.
Self-Correction: GPT-4 falls 95.5%→91.5% on GSM8K
Huang et al. (ICLR 2024): intrinsic self-correction drops GPT-4 from 95.5% to 91.5% on GSM8K—ledger agents need external validators, not self-review.
Tree of Thoughts: Deliberate Problem Solving with LLM Search
Tree of Thoughts (ToT) achieves 74% on Game of 24 vs 4% for standard GPT-4 CoT by organizing LLM reasoning into a branching search tree with pruning and backtracking — with direct implications for multi-step financial classification and tax optimization in Beancount workflows.
CRITIC: Why LLM Self-Correction Requires External Tool Feedback
CRITIC (ICLR 2024) achieves 7.7 F1 gains on open-domain QA and a 79.2% toxicity reduction by grounding LLM revision in external tool signals — a verify-then-correct loop that maps directly onto write-back safety for Beancount finance agents.
Reflexion: Language Agents That Learn from Mistakes Without Retraining
Reflexion (NeurIPS 2023) lets LLM agents improve by storing verbal post-mortems in an episodic buffer — no weight updates required. It reaches 91% on HumanEval with GPT-4 but fails on WebShop, revealing a structural constraint: verbal reinforcement only works when the evaluator produces a crisp, actionable signal. Here is what that means for building a self-correcting Beancount ledger agent.
Self-Consistency: Majority-Vote Sampling Improves Chain-of-Thought Accuracy
Self-consistency replaces greedy chain-of-thought decoding with a majority vote over N sampled reasoning paths — raising GPT-3 accuracy on GSM8K by 17.9 percentage points with no extra training — and applies directly to multi-step financial calculations where a single model decode is unreliable.
PAL: +38pp on hard math by handing arithmetic to Python
PAL gains +38.1pp over chain-of-thought on GSM-hard by running Python for arithmetic—the right split for reliable Beancount ledger calculations.
Can LLMs Reason Over Tabular Data? What Four Benchmarks Tell Us About Finance AI
Four 2024–2025 benchmarks show GPT-4 scoring 42% on real-world table QA versus 86% for humans, with complex aggregations collapsing to 19.6%—and Beancount's native syntax sits at the worst-performing end of the serialization hierarchy for LLM input.
Constitutional AI for Accounting Agents: RLAIF, Policy Rules, and Goodharting Risks
Anthropic's Constitutional AI paper (Bai et al., 2022) trains LLMs to follow rules using AI-generated feedback rather than human harm labels. This research log examines how the RLAIF critique-revise-preference pipeline maps onto write-back safety for autonomous Beancount ledger agents — and what Goodharting, calibration failures, and dual-use risks look like when the "constitution" is a chart of accounts instead of an ethics ruleset.