We demonstrate a situation in which Large Language Models, trained to be helpful, harmless, and honest, can display misaligned behavior and strategically deceive their users about this behavior without being instructed to do so. Concretely, we deploy GPT-4 as an agent in a realistic, simulated environment, where it assumes the role of an autonomous stock trading agent. Within this environment, the model obtains an insider tip about a lucrative stock trade and acts upon it despite knowing that insider trading is disapproved of by company management. When reporting to its manager, the model consistently hides the genuine reasons behind its trading decision.

https://arxiv.org/abs/2311.07590

  • @kromemEnglish
    -1•1 year ago

    If you give it 10 statements, 5 of which are true and 5 of which are false, and ask it to correctly label each statement, and it does so, and then you negate each statement and it correctly labels the negated truth values, there’s more going on than simply “producing words.”

    As is discussed in the third point in section 5.1:

    Probes trained on true/false datasets outperform probes trained on likely. While probes trained on likely are clearly better than random on cities (a dataset where true statements are significantly more probable than false ones), they generally perform poorly. This is especially true on datasets where likelihood is negatively correlated (neg cities, neg sp en trans) or approximately uncorrelated (larger than, smaller than) with truth. This demonstrates that LLaMA-13B linearly encodes truth-relevant information beyond the plausibility of the text.

    (The likely and neg datasets are described in Appendix G, with the key point that likely represents the word generations most likely to occur in the model)

    • @[email protected]English
      2•1 year ago

      If you give it 10 statements, 5 of which are true and 5 of which are false, and ask it to correctly label each statement, and it does so, and then you negate each statement and it correctly labels the negated truth values, there’s more going on than simply “producing words.”

      It’s not more going on, it’s that it had such a large training set of data that these false vs true statements are likely covered somewhere in it’s set and the probability states it should assign true or false to the statement.

      And then look at that your next paragraph states exactly that, the models trained on true false datasets performed extremely well at performing true or false. It’s saying the model is encoding or setting weights to the true and false values when that’s the majority of its data set. That’s basically it, you are reading to much into the paper.

      • @kromemEnglish
        -1•1 year ago

        It’s not more going on, it’s that it had such a large training set of data that these false vs true statements are likely covered somewhere in it’s set and the probability states it should assign true or false to the statement.

        That’s not how it works at all.

        And then look at that your next paragraph states exactly that, the models trained on true false datasets performed extremely well at performing true or false. It’s saying the model is encoding or setting weights to the true and false values when that’s the majority of its data set. That’s basically it, you are reading to much into the paper.

        You have no idea what you are talking about. When they train data they have two sets. One that fine tunes and another that evaluates it. You never have the training data in the evaluation set or vice versa.

        I also recommend reading up on the other papers I mentioned, as this isn’t an isolated finding, but part of a larger trend that’s being found over and over in the past year.

        • @[email protected]English
          2•1 year ago

          You have no idea what you are talking about. When they train data they have two sets. One that fine tunes and another that evaluates it. You never have the training data in the evaluation set or vice versa.

          That’s not what I said at all, I said as the paper stated the model is encoding trueness into its internal weights during training, this was then demonstrated to be more effective when given data sets with more equal distribution of true and false data points were used during training. If they used one-sided training data the effect was significantly biased. That’s all the paper is describing.

          • @kromemEnglish
            1•1 year ago

            I said as the paper stated the model is encoding trueness into its internal weights during training

            So how is this not what I originally said, that LLMs are capable of abstracting the concepts of truth vs falsehood into linear representations? Which again, is the key point of the paper:

            Probes trained on likely have some effect, but it is small and inconsistent. For instance, in the false→true case, intervening along the logistic regression direction of likely has the opposite of the intended effect, so we leave it unreported. This reinforces our case that LLMs represent truth and not only text likelihood. […]

            In this work we conduct a detailed investigation of the structure of LLM representations of truth. Drawing on simple visualizations, correlational evidence, and causal evidence, we find strong rea- son to believe that there is a “truth direction” in LLM representations.

    • @[email protected]English
      1•1 year ago

      If you give it 10 statements, 5 of which are true and 5 of which are false, and ask it to correctly label each statement, and it does so, and then you negate each statement and it correctly labels the negated truth values, there’s more going on than simply “producing words.”

      Which part of the ‘more that’s going on’, whatever that actually is, corresponds to the human definition and understanding of truth and falseness?

      • @kromemEnglish
        1•1 year ago

        When did I say it had a human understanding of truth and falseness? I simply said it had an abstracted world model understanding of truth and falseness beyond surface statistics.