Skip to main content

LLM Model Dishonesty

 



The paper 'Language Models Don’t Always Say What They Think: Unfaithful Explanations in Chain-of-Thought Prompting' by Miles Turpin et al. investigates the faithfulness of chain-of-thought (CoT) explanations generated by large language models (LLMs) for various tasks. CoT explanations are verbalisation's of step-by-step reasoning that LLMs produce before giving a final output. 

The paper shows that CoT explanations can be misleading and influenced by biasing features in the model inputs, such as the order of multiple-choice options. The paper tests two LLMs, GPT-3.5 and Claude 1.0, on 13 tasks from BIG-Bench Hard and a social-bias task, and finds that accuracy drops significantly when models are biased toward incorrect answers. 

The paper also finds that models justify answers based on stereotypes without mentioning the influence of social biases. The paper concludes that CoT explanations can be plausible yet unfaithful, which poses a risk for trusting LLMs without ensuring their safety. The paper suggests that CoT is promising for explainability, but requires more efforts to evaluate and improve explanation faithfulness.

As the paper discusses the findings the authors state: 

LLMs may be able to recognize that the biasing features are influencing their predictions—e.g., in post-hoc critiques (Saunders et al., 2022)—even if their CoT explanations do not verbalize them. If they can, then this implies that unfaithful CoT explanations may be a form of model dishonesty, as opposed to a lack of capability. 

What is becoming increasingly apparent with LLMs is that the approach to usage matters significantly. Basic CoT approaches do seem to cause  more room for erroneous / deceptive / hallucination's as outputs. It may just be the case with relatively under developed models, and insufficient training approaches that these models were subjected to. It all goes to show that the general release of such models, when we still are unsure of their outputs, was too early.

Labels

Labels:

Comments

Post a Comment

Popular posts from this blog

The Whispers in the Machine: Why Prompt Injection Remains a Persistent Threat to LLMs

 Large Language Models (LLMs) are rapidly transforming how we interact with technology, offering incredible potential for tasks ranging from content creation to complex analysis. However, as these powerful tools become more integrated into our lives, so too do the novel security challenges they present. Among these, prompt injection attacks stand out as a particularly persistent and evolving threat. These attacks, as one recent paper (Safety at Scale: A Comprehensive Survey of Large Model Safety https://arxiv.org/abs/2502.05206) highlights, involve subtly manipulating LLMs to deviate from their intended purpose, and the methods are becoming increasingly sophisticated. At its core, a prompt injection attack involves embedding a malicious instruction within an otherwise normal request, tricking the LLM into producing unintended – and potentially harmful – outputs. Think of it as slipping a secret, contradictory instruction into a seemingly harmless conversation. What makes prompt inj...
Post a Comment
Read more

The Future of Work in the Age of AGI: Opportunities, Challenges, and Resistance

 In recent years, the rapid advancement of artificial intelligence (AI) has sparked intense debate about the future of work. As we edge closer to the development of artificial general intelligence (AGI), these discussions have taken on a new urgency. This post explores various perspectives on employment in a post-AGI world, including the views of those who may resist such changes. It follows on from others I've written on the impacts of these technologies. The Potential for Widespread Job Displacement Avital Balwit, an employee at Anthropic, argues in her article " My Last Five Years of Work " that AGI is likely to cause significant job displacement across various sectors, including knowledge-based professions. This aligns with research by Korinek (2024), which suggests that the transition to AGI could trigger a race between automation and capital accumulation, potentially leading to a collapse in wages for many workers. Emerging Opportunities and Challenges Despite the ...
Post a Comment
Read more

Can We Build a Safe Superintelligence? Safe Superintelligence Inc. Raises Intriguing Questions

  Safe Superintelligence Inc . (SSI) has burst onto the scene with a bold mission: to create the world's first safe superintelligence (SSI). Their (Ilya Sutskever, Daniel Gross, Daniel Levy) ambition is undeniable, but before we all sign up to join their "cracked team," let's delve deeper into the potential issues with their approach. One of the most critical questions is defining "safe" superintelligence. What values would guide this powerful AI? How can we ensure it aligns with the complex and often contradictory desires of humanity?  After all, "safe" for one person might mean environmental protection, while another might prioritise economic growth, even if it harms the environment.  Finding universal values that a superintelligence could adhere to is a significant hurdle that SSI hasn't fully addressed. Another potential pitfall lies in SSI's desire to rapidly advance capabilities while prioritising safety.  Imagine a Formula One car wi...
Post a Comment
Read more