5 ms·
LLM Output Drift in Financial Workflows: Validation and Mitigation (arXiv)
- raffisk 11mo agoEmpirical study on LLM output consistency in regulated financial tasks (RAG, JSON, SQL). Governance focus: Smaller models (Qwen2.5-7B, Granite-3-8B) hit 100% determinism at T=0.0, passing audits (FSB/BIS/CFTC), vs. larger like GPT-OSS-120B at 12.5%. Gaps are huge (87.5%, p<0.0001, n=16) and survive multiple-testing corrections. Caveat: Measures reproducibility (edit distance), not full accuracy—determinism is necessary for compliance but needs semantic checks (e.g., embeddings to ground truth). Includes harness, invariants (±5%), and attestation. Thoughts on inverse size-reliability? Planning follow-up with accuracy metrics vs. just repro.
- colechristensen 11mo agoOutputs not being deterministic with temperature = 0 doesn't match my understanding of what "temperature" meant, I thought the definition of T=0 was determinism. Is this perhaps inference implementation details somehow introducing randomness?
- kakugawa 11mo agoDefeating Nondeterminism in LLM Inference https://news.ycombinator.com/item?id=45200925 https://news.ycombinator.com/item?id=45200925 https://thinkingmachines.ai/blog/defeating-nondeterminism-in-llm-inference/ https://thinkingmachines.ai/blog/defeating-nondeterminism-in... > As it turns out, our request’s output does depend on the parallel user requests. Not because we’re somehow leaking information across batches — instead, it’s because our forward pass lacks “batch invariance”, causing our request’s output to depend on the batch size of our forward pass. tl;dr: the way inference is batched introduces non-determinism.
- doctorpangloss 11mo ago“Determinism is necessary for compliance” Says who? The stuff you comply with changes in real time. How’s that for determinism?
- raffisk 11mo agoAuthor here—fair point, regs are a moving target . But FSB/BIS/CFTC explicitly require reproducible outputs for audits (no random drift in financial reports). Determinism = traceability, even when rules update at the very least Most groups I work with stick to traditional automation/rules systems, but top-down mandates are pushing them toward frontier models for general tasks—which then get plugged into these workflows. A lot stays in sandbox, but you'd be surprised what's already live in fin services. The authorities I cited (FSB/BIS/CFTC) literally just said last month AI monitoring is "still at early stage" cc https://www.fsb.org/2024/11/the-financial-stability-implications-of-artificial-intelligence/ https://www.fsb.org/2024/11/the-financial-stability-implicat... Curious how you'd tackle that real-time changing reg?
- raffisk 11mo ago* https://www.fsb.org/2025/10/monitoring-adoption-of-artificial-intelligence-and-related-vulnerabilities-in-the-financial-sector/ https://www.fsb.org/2025/10/monitoring-adoption-of-artificia... This was the link I meant from Oct ‘25 reiterating early stages of AI monitoring
- nomel 11mo agoAlso, what happens if you add a space to the end of the prompt? Or write a 12.00 to 12.000?
- raffisk 11mo agoGood q—spacing could mess with tokenization, untested but def plausible. Worth a quick test on the setup - through the code for the fin svcs harness for tinkering / testing diff prompts/model arch’s based on feedback https://github.com/ibm-client-engineering/output-drift-financial-llms https://github.com/ibm-client-engineering/output-drift-finan...
- ulrashida 11mo agoPlease give an example of a statutory compliance item that "changes in real time". That's not the way regulations work. Your compliance is measured against a fixed version of legislation.
- throwdbaaway 11mo agoIt is the reasoning. During the reasoning process, the top few tokens have very similar or even same logprobs. With gpt-oss-120b, you should be able to get deterministic output by turning off reasoning, e.g. by appending: {"role": "assistant", "content": "<think></think>"} Of course, the model will be less capable without reasoning.
- raffisk 11mo agoGood call—reasoning token variance is likely a factor, esp with logprob clustering at T=0. Your <think></think> workaround would work, but we need reasoning intact for financial QA accuracy. Also the mistral medium model we tested had ~70% deterministic outputs across the 16 runs for the text to sql gen and summarization in json tasks- and it had reasoning on. Llama 3.3 70b started to degrade and doesn’t have reasoning. But it’s a relevant variable to consider
- measurablefunc 11mo agoThis is b/c these things are Markov chains. You can not expect consistent results & outputs.
- SrslyJosh 11mo agoUsing an LLM for a "financial workflow" makes as much sense as integrating one with Excel. But who needs correct results when you're just working with money, right? ¯\_(ツ)_/¯
- mirekrusin 11mo agoHumans are non deterministic yet they use excel, work with financial workflows and deal with the money.
- Terr_ 11mo ago"Humans make math errors, yet they do math anyway, therefore this calculator that makes errors is also OK." What do you call the fallacy where the universe is imperfect, therefore nobody can have higher standards for anything? Mankind has spent literal centuries observing deficiencies and faults in human bookkeeping and calculation, constantly trying to improve it with processes and machinery. There's no good reason to suddenly stop caring about those issues simply because the latest proposal is marketed as "AI".
- mirekrusin 11mo agoIt can interact with deterministic and provable systems just fine.
- thfuran 11mo agoAnd because one system that aims to achieve deterministic operation can’t quite perfectly do so, we might as well abandon any attempt at determinism?
- measurablefunc 11mo ago
- 34679 11mo agoDon't use LLMs for financial workflows. Use them to create software for financial workflows. Software doesn't "drift".
- wild_pointer 11mo agoLLM-created software might