How LLMs Interpret Unannotated Datasets

The "black box" nature of LLMs poses significant challenges for regulation and ensuring safety. Due to their opaque and complex internal workings, it is often not clear how these models arrive at specific answers or why they generate certain outputs. This lack of transparency complicates efforts to establish robust regulatory frameworks, as regulators find it difficult to assess compliance with ethical and legal standards, including privacy and fairness. Furthermore, without a clear understanding of how answers are generated, users may question the reliability and trustworthiness of the responses they receive. This uncertainty can deter wider adoption and reliance on LLMs. This study (https://lnkd.in/efjmvwiw) aims to address some of these issues by introducing CausalBench which is designed to address the limitations of existing causal evaluation methods by enhancing the complexity and diversity of the data, tasks, and prompt formats used in the assessments. The purpose of CausalBench is to test and understand the limits of LLMs in identifying and reasoning about causality particularly how well they can perform under conditions that mimic real-world examples. Using CausalBench, the authors then evaluated 19 leading LLMs on their capability to discern direct and indirect correlations, construct causal skeletons, and identify explicit causality from structured and unstructured data. Here are the key takeaways: • 𝗦𝗲𝗻𝘀𝗶𝘁𝗶𝘃𝗶𝘁𝘆 𝘁𝗼 𝗗𝗮𝘁𝗮𝘀𝗲𝘁 𝗦𝗰𝗮𝗹𝗲: LLMs are capable of recognizing direct correlations in smaller datasets, but their performance declines with larger, more complex datasets, particularly in detecting indirect correlations. This indicates a need for models trained on larger and more complex network structures. • 𝗣𝗲𝗿𝗳𝗼𝗿𝗺𝗮𝗻𝗰𝗲 𝗘𝗱𝗴𝗲 𝗼𝗳 𝗖𝗹𝗼𝘀𝗲𝗱-𝘀𝗼𝘂𝗿𝗰𝗲 𝗟𝗟𝗠𝘀: Closed-source LLMs like GPT3.5-Turbo and GPT4 outperform open-source models in causality-related tasks, suggesting that the extensive training data and diverse datasets used for these models enhance their ability to handle complex causal queries. • 𝗜𝗺𝗽𝗮𝗰𝘁 𝗼𝗳 𝗣𝗿𝗼𝗺𝗽𝘁 𝗗𝗲𝘀𝗶𝗴𝗻: The effectiveness of LLMs varies with different prompt formats, with combinations of variable names with structured data or background knowledge proving particularly beneficial. The development of comprehensive benchmarks like CausalBench is pivotal in demystifying the "black box" nature of LLMs. This enhanced transparency aids in complex reasoning tasks, guiding the selection of appropriate models for specific applications based on empirical performance data. Additionally, a more granular understanding of LLM capabilities and behaviors facilitates more effective regulation and risk management, addressing both ethical and practical concerns in deploying these models in sensitive or high-stakes environments.

😅 If not properly configured, LLM judges can cause more trouble than they solve. LLM judges are quickly becoming a go-to for evaluating LLM results, cutting down on human effort. However, they must be carefully configured, either through training, proper cues, or human annotations. Here’s a nice paper from Meta that shows how to achieve this using only synthetic training data, without relying on human annotations. Some Insights: ⛳ The paper uses unlabeled instructions and prompting to generate synthetic preference pairs, where one response is intentionally made inferior to the other. ⛳ An LLM is then used to generate reasoning traces and judgments for these pairs, creating labeled data from the synthetic examples. ⛳ This labeled data is used to retrain the LLM-as-a-Judge, with the process repeated in cycles to progressively improve the model’s evaluation capabilities. ⛳ On the Llama-3-70B-Instruct model, the approach obtains accuracy on RewardBench from 75.4 to 88.7 (with majority vote) or 88.3 (without majority vote). The method matches or even outperforms traditional reward models trained on human-annotated data, demonstrating the potential of using synthetic data for model evaluation without relying on human input. Link: https://lnkd.in/eRhF4ykx

One of our core areas of work is curating datasets, and this also represents our use of LLMs for label extraction. When we begun labeling our Emory CXR dataset, we extracted both CheXpert labels and LLM labels. At that point we needed to make a decision - will we keep chasing every new LLM update (while remaining cost conscious) to update our labels or what was going to be our guide? We share our experience in this paper - https://rdcu.be/ekNCl led by Bardia Khosravi. We developed #Radprompter to help with tracking and versioning our prompts with new updates. We found - 1. LLM-based labeling outperformed the CheXpert labeler, with the best LLM achieving 95% sensitivity for fracture detection versus CheXpert’s 51%. 2. Larger models showed better sensitivity, while chain-of-thought (COT) prompting had variable effects. 3. Image classifiers showed resilience to labeling noise when tested externally. Simply stated - We can rely on using the smallest models with good performance. Be careful with COT as the models tend to "overthink" resulting in inaccuracies. For downstream task model training, LLM extracted labels are almost as good as human extracted labels - which means you can direct your resources to the test dataset. We found that the models trained on these labels were very robust to label noise.